Photovoltaic power generation data correction method, photovoltaic power generation prediction method and system

Abstract

The invention belongs to the field of photovoltaic power generation, and particularly relates to a photovoltaic power generation data correction method, a photovoltaic power generation prediction method and a photovoltaic power generation prediction system. According to the method, the historical output data of the similar days of the similar power stations and the power data points corresponding to the to-be-corrected data are selected, and the correction model corresponding to the to-be-corrected data is trained, which is equivalent to correction by using the data of the similar days of the power stations; and the historical output data of the similar power stations in the same area are utilized to carry out auxiliary correction on the power stations, that is, the correction model contains two data correction references of the similar power stations and similar days at the same time, so that the correction of the to-be-corrected data can be restrained from two dimensions, and the correction effect of the correction model is improved. The corrected historical output data are closer to the actual situation, and the correction precision can be greatly improved, so that the credibility of data correction is improved.

Description

Technical Field

[0001] The present invention belongs to the field of photovoltaic power generation, and in particular relates to a photovoltaic power generation data correction method, a photovoltaic power generation prediction method and a system. Background Art

[0002] At present, under the guidance of the national dual carbon goals, the process of clean energy replacing non-renewable energy is accelerating. my country's solar energy resources are relatively abundant. More than 2 / 3 of the country's land area has an annual sunshine hour of more than 2,000 hours, and the annual radiation is 5,000MJ / m 2 As a kind of inexhaustible renewable clean energy, the solar energy absorbed by the earth's surface each year is equivalent to about 2,000 times the energy currently obtained by burning various traditional fossil fuels on the earth each year. Photovoltaic power generation is one of the core forms of new energy power generation and the main way to utilize solar energy. It is usually composed of power stations with batteries that can be dispatched in real time and power stations without batteries that have poor dispatchability.

[0003] At present, in the management process of photovoltaic power generation systems, the results of photovoltaic power generation are usually predicted, which can provide accurate data support for the operation management of power stations and grid operation dispatching. Some local states can be controlled from multiple aspects to ensure the safety and reliability of the operation of the entire power system; however, when there is sudden damage to the equipment of the photovoltaic power station, temporary congestion in communication and bit errors during transmission, the monitoring and storage of real-time data by the monitoring equipment will be abnormal, resulting in bad data information. When the historical data collected by the photovoltaic site is inaccurate, it will also cause great unreliability in the subsequent photovoltaic power generation prediction, resulting in inaccurate power generation prediction of the photovoltaic site, and unable to provide more accurate prediction information for dispatching, which affects the operation management of the power station and grid operation dispatching. Summary of the invention

[0004] The purpose of the present invention is to provide a photovoltaic power generation data correction method, a photovoltaic power generation prediction method and a system, which are used to solve the problem in the prior art that problems arise when collecting real-time data from photovoltaic stations, resulting in inaccurate prediction results of photovoltaic power generation.

[0005] In order to achieve the above object, the present invention provides a photovoltaic power generation data correction method, comprising the following steps:

[0006] Obtain historical output data of each photovoltaic power station in a set geographical area where the photovoltaic power station corresponding to the data to be corrected is located; obtain similarities between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected, and determine similar power stations to the photovoltaic power station corresponding to the data to be corrected according to the similarities;

[0007] Obtaining meteorological information data corresponding to the collection day of the data to be corrected, and determining a similar day corresponding to the collection day in similar power stations; the similarity between the meteorological information data of the similar day corresponding to the collection day in similar power stations and the meteorological information data corresponding to the collection day is greater than a set threshold;

[0008] According to the historical output data collected on the similar day of the corresponding collection day in similar power stations and the meteorological information data on the similar day of the corresponding collection day, the correction model corresponding to the data to be corrected is trained;

[0009] After the training of the correction model corresponding to the data to be corrected is completed, the normal historical output data and meteorological information data corresponding to the collection day of the data to be corrected are input into the correction model to obtain the corrected photovoltaic power generation data.

[0010] The beneficial effect of the above technical solution is that on the basis of using the data of similar days of the power station for correction, the historical output data of similar power stations in the same area can be used to assist in the correction of the power station, that is, the correction model also includes two data correction references of similar power stations and similar days. Therefore, the correction of the data to be corrected can be constrained from two dimensions, thereby improving the correction effect of the correction model, making the corrected historical output data closer to the actual situation, and being able to greatly improve the correction accuracy, thereby improving the credibility of the data correction.

[0011] Further, similarities between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected are obtained, and a method of determining similar power stations of the photovoltaic power station corresponding to the data to be corrected according to the similarities is as follows:

[0012] In the historical output data of photovoltaic power station A corresponding to the data to be corrected, the output power of photovoltaic power station A corresponding to the data to be corrected is:

[0013] P A = {P A1 ,P A2 ,...,P An}

[0014] In the historical output data of other photovoltaic power stations in the same set geographical area, the output power of other photovoltaic power stations B, C and D are:

[0015] P B = {P B1 ,P B2 ,...,P Bn}

[0016] P C = {P C1 ,P C2 ,...,P Cn}

[0017] P D = {P D1 ,P D2 ,...,P Dn}

[0018] The similarity S between the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area is A,B for:

[0019] S A,B =λd A,B +γr A,B

[0020] Where λ and γ are d A,B and r A,B The weight coefficient is: the photovoltaic power station whose similarity is greater than the set similarity threshold and is located in the same set geographical area as the photovoltaic power station A corresponding to the data to be corrected as a similar power station, or the photovoltaic power station A corresponding to the data to be corrected is located in the same set geographical area and has the greatest similarity to the photovoltaic power station A corresponding to the data to be corrected as a similar power station, d A,B and r A,B The calculation formula is as follows:

[0021]

[0022]

[0023] Where n is the number of indicators of the output power sample, k represents the kth indicator, and p ak and p bk They are respectively the k-th point of power output of the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area.

[0024] The beneficial effect of the above technical solution is: using a similarity statistic that combines numerical proximity and curve shape proximity to judge similar power stations, it can not only measure the numerical proximity, but also correctly distinguish the shape trend of the curve, thereby ensuring the reliability of the determined similar power stations.

[0025] Furthermore, according to the historical output data collected on the similar day of the corresponding collection day in similar power stations and the meteorological information data on the similar day of the corresponding collection day, the method of training the correction model corresponding to the data to be corrected is as follows:

[0026] 1) From the historical output data collected on a similar day of the corresponding collection day in similar power stations, select n output power data points [p(i), p(i+n-1)] corresponding to the data point before the data to be corrected as samples, combine the meteorological information data of the similar day of the corresponding collection day, predict the first output power data point p(i+n) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n) in the historical output data collected on the similar day to train the first data point of the data to be corrected in the correction model corresponding to the data to be corrected; wherein n is an integer greater than 1;

[0027] 2) Taking n-1 output power data points [p(i+1), p(i+n-1)] corresponding to the data point before the data to be corrected and the predicted output power data point p(i+n) as samples, as well as the meteorological information data of a similar day to the corresponding collection day, predict the second output power data point p(i+n+1) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n+1) in the historical output data collected on the similar day to train the second data point of the data to be corrected in the correction model corresponding to the data to be corrected;

[0028] 3) Repeat the prediction and training process of step 2) above until all output power data points corresponding to the data to be corrected have been predicted and all data points of the data to be corrected in the correction model corresponding to the data to be corrected have been trained, thus completing the training of the correction model corresponding to the data to be corrected.

[0029] Furthermore, the meteorological information data includes weather type, maximum temperature, minimum temperature and average temperature.

[0030] The present invention also provides a photovoltaic power generation data correction system, comprising a processor, wherein the processor is used to execute program instructions to implement the above photovoltaic power generation data correction method.

[0031] The photovoltaic power generation data correction system can achieve the same beneficial effects as the photovoltaic power generation data correction method described above.

[0032] The present invention also provides a photovoltaic power generation prediction method, comprising the following steps:

[0033] Acquire and pre-process photovoltaic power generation data, wherein the photovoltaic power generation data includes historical output data of each photovoltaic power station, atmospheric temperature data, and solar irradiation data;

[0034] Obtaining historical output data of each photovoltaic power station in a set geographical area where the photovoltaic power station corresponding to the preprocessed data to be corrected is located; obtaining similarities between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected, and determining similar power stations to the photovoltaic power station corresponding to the data to be corrected according to the similarities;

[0035] Obtaining meteorological information data corresponding to the collection day of the data to be corrected, and determining a similar day corresponding to the collection day in similar power stations; the similarity between the meteorological information data of the similar day corresponding to the collection day in similar power stations and the meteorological information data corresponding to the collection day is greater than a set threshold;

[0036] According to the historical output data collected on the similar day of the corresponding collection day in similar power stations and the meteorological information data on the similar day of the corresponding collection day, the correction model corresponding to the data to be corrected is trained;

[0037] After the correction model training corresponding to the data to be corrected is completed, the normal historical output data and meteorological information data corresponding to the collection day of the data to be corrected are input into the correction model to obtain the corrected historical output data;

[0038] The corrected historical output data and the pre-processed atmospheric temperature data and solar irradiation data are input into the prediction model to predict the photovoltaic power generation and obtain the prediction results.

[0039] The beneficial effect of the above technical solution is: on the basis of using the data of similar days of the power station for correction, the historical output data of similar power stations in the same area are also used to assist in the correction of the power station, that is, the correction model simultaneously includes two data correction references of similar power stations and similar days. Therefore, the correction of the data to be corrected can be constrained from two dimensions, thereby improving the correction effect of the correction model, making the corrected historical output data closer to the actual situation, and being able to greatly improve the correction accuracy, thereby improving the credibility of the data correction. Therefore, using the corrected data for photovoltaic power generation prediction can improve the accuracy of the prediction results.

[0040] Further, similarities between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected are obtained, and a method of determining similar power stations of the photovoltaic power station corresponding to the data to be corrected according to the similarities is as follows:

[0041] In the historical output data of photovoltaic power station A corresponding to the data to be corrected, the output power of photovoltaic power station A corresponding to the data to be corrected is:

[0042] P A = {P A1 ,P A2 ,...,P An}

[0043] In the historical output data of other photovoltaic power stations in the same set geographical area, the output power of other photovoltaic power stations B, C and D are:

[0044] P B = {P B1 ,P B2 ,...,P Bn}

[0045] P C = {P C1 ,P C2 ,...,P Cn}

[0046] P D = {P D1 ,P D2 ,...,P Dn}

[0047] The similarity S between the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area is A,B for:

[0048] S A,B =λd A,B +γr A,B

[0049] Where λ and γ are d A,B and r A,B The weight coefficient is: the photovoltaic power station whose similarity is greater than the set similarity threshold and is located in the same set geographical area as the photovoltaic power station A corresponding to the data to be corrected as a similar power station, or the photovoltaic power station A corresponding to the data to be corrected is located in the same set geographical area and has the greatest similarity to the photovoltaic power station A corresponding to the data to be corrected as a similar power station, d A,B and r A,B The calculation formula is as follows:

[0050]

[0051]

[0052] Where n is the number of indicators of the output power sample, k represents the kth indicator, and p ak and pbk They are respectively the k-th point of power output of the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area.

[0053] The beneficial effect of the above technical solution is: using a similarity statistic that combines numerical proximity and curve shape proximity to judge similar power stations, it can not only measure the numerical proximity, but also correctly distinguish the shape trend of the curve, thereby ensuring the reliability of the determined similar power stations.

[0054] Furthermore, according to the historical output data collected on the similar day of the corresponding collection day in similar power stations and the meteorological information data on the similar day of the corresponding collection day, the method of training the correction model corresponding to the data to be corrected is as follows:

[0055] 1) From the historical output data collected on a similar day of the corresponding collection day in similar power stations, select n output power data points [p(i), p(i+n-1)] corresponding to the data point before the data to be corrected as samples, combine the meteorological information data of the similar day of the corresponding collection day, predict the first output power data point p(i+n) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n) in the historical output data collected on the similar day to train the first data point of the data to be corrected in the correction model corresponding to the data to be corrected; wherein n is an integer greater than 1;

[0056] 2) Taking n-1 output power data points [p(i+1), p(i+n-1)] corresponding to the data point before the data to be corrected and the predicted output power data point p(i+n) as samples, as well as the meteorological information data of a similar day to the corresponding collection day, predict the second output power data point p(i+n+1) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n+1) in the historical output data collected on the similar day to train the second data point of the data to be corrected in the correction model corresponding to the data to be corrected;

[0057] 3) Repeat the prediction and training process of step 2) above until all output power data points corresponding to the data to be corrected have been predicted and all data points of the data to be corrected in the correction model corresponding to the data to be corrected have been trained, thus completing the training of the correction model corresponding to the data to be corrected.

[0058] Furthermore, the meteorological information data includes weather type, maximum temperature, minimum temperature and average temperature.

[0059] The present invention also provides a photovoltaic power generation prediction system, comprising a processor, wherein the processor is used to execute program instructions to implement the photovoltaic power generation prediction method as described above.

[0060] The photovoltaic power generation prediction system can achieve the same beneficial effects as the photovoltaic power generation prediction method described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flowchart of the photovoltaic power generation prediction method in the photovoltaic power generation prediction method embodiment of the present invention;

[0062] Figure 2 It is a schematic diagram of comparing the historical output data before correction and the actual data in the photovoltaic power generation prediction method embodiment of the present invention;

[0063] Figure 3 It is a schematic diagram of comparing the corrected historical output data and the actual data in the embodiment of the photovoltaic power generation prediction method of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0065] Photovoltaic power generation prediction method embodiment

[0066] This embodiment provides a photovoltaic power generation prediction method. Figure 1 , the photovoltaic power generation prediction method comprises the following steps:

[0067] Step 1, obtain photovoltaic power generation data and perform preprocessing; the photovoltaic power generation data includes historical output data of each photovoltaic power station, atmospheric temperature data and solar irradiation data; the historical output data of each photovoltaic power station is the historical output power data of each photovoltaic power station.

[0068] The preprocessing of photovoltaic power generation data of multiple sites includes removing noise and performing data normalization preprocessing. The wavelet transform is used to reduce noise on the data, which can well retain the peak value and mutation part of the useful signal required in the original signal. The high-frequency part of the required signal and the high-frequency interference caused by noise can be effectively distinguished through the wavelet transform. In this embodiment, the historical photovoltaic power generation data within the time range from one day to N days before the prediction date is selected by installing a power measurement device, a thermometer and an irradiance meter in each distributed photovoltaic power generation system, and the output power data, temperature data and irradiance intensity data of the distributed photovoltaic power generation system are collected at a sampling rate of M sampling points per day, and then the output power data, atmospheric temperature data and irradiance intensity data are respectively normalized according to the following formulas:

[0069]

[0070] Among them, P(j) and are the photovoltaic power generation data at time j before and after normalization, j = 1, 2, ..., M; P min and P max are the minimum and maximum values ​​of photovoltaic power generation data respectively; among them, the collected output power data, temperature data and irradiation intensity data are all collected from historical data within the time range of one day before the day to be predicted to the previous N days.

[0071] Step 2, obtaining the historical output data of each photovoltaic power station in the set geographical area where the photovoltaic power station corresponding to the pre-processed data to be corrected is located; according to the historical output data of each photovoltaic power station, obtaining the similarity between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected, and determining the similar power stations of the photovoltaic power station corresponding to the data to be corrected according to the similarity; in order to measure the numerical proximity and correctly distinguish the shape trend of the curve, a new similarity statistic S that combines the value proximity and the shape proximity is adopted. A,B , wherein the angle cosine calculation method is used in the analysis of shape correlation; specifically, in this embodiment, the method of determining similar power stations is as follows:

[0072] In the historical output data of photovoltaic power station A corresponding to the data to be corrected, the output power of photovoltaic power station A corresponding to the data to be corrected is:

[0073] P A = {P A1 ,P A2 ,...,P An}

[0074] In the historical output data of other photovoltaic power stations B, C and D in the same set geographical area, the output power of other photovoltaic power stations B, C and D are:

[0075] P B = {P B1 ,P B2 ,...,P Bn}

[0076] P C = {P C1 ,P C2 ,...,P Cn}

[0077] P D = {P D1 ,P D2 ,...,P Dn}

[0078] The similarity S between the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area is A,B for:

[0079] S A,B =λd A,B +γr A,B

[0080] Where λ and γ are d A,B and r A,B The weight coefficient of ; the appropriate weight coefficient can screen out samples with more accurate similarity, thereby improving the restoration accuracy. In this embodiment, λ and γ are both set to 0.5;

[0081] The photovoltaic power station with a similarity greater than the set similarity threshold and located in the same set geographical area as the photovoltaic power station A corresponding to the data to be corrected is regarded as a similar power station, or the photovoltaic power station with the greatest similarity to the photovoltaic power station A corresponding to the data to be corrected is regarded as a similar power station. A,B and r A,B The calculation formula is as follows:

[0082]

[0083]

[0084] Where n is the number of indicators of the output power sample, k represents the kth indicator, and p ak and p bk are the kth points of the power output of the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area; in this embodiment, the cosine of the angle r is used A,B As a measure of shape similarity, the larger the value, the closer the output power curves of the two PV power stations are to each other.

[0085] Obtain meteorological information data corresponding to a collection day of the data to be corrected, and after determining similar power stations, determine a similar day in the similar power stations that corresponds to the collection day; in this embodiment, the similarity between the meteorological information data of a similar day in the similar power stations that corresponds to the collection day and the meteorological information data corresponding to the collection day is greater than a set threshold; in other embodiments, a collection day in a similar power station that has the same weather type as the collection day of the data to be corrected may also be used as a similar day.

[0086] After determining the similar day corresponding to the collection day in the similar power station, the correction model corresponding to the data to be corrected is trained according to the historical output data collected on the similar day corresponding to the collection day in the similar power station and the meteorological information data on the similar day corresponding to the collection day (in this embodiment, the data to be corrected mainly refers to the data to be corrected corresponding to the local data defects). The specific training process is as follows:

[0087] 1) From the historical output data collected on similar days of the corresponding collection day in similar power stations, select n output power data points [p(i), p(i+n-1)] corresponding to the data point before the data to be corrected as samples, combine the meteorological information data of the similar days of the corresponding collection day, predict the first output power data point p(i+n) corresponding to the data point of the data to be corrected in the historical output data collected on these similar days, and combine the actual value of the output power data point p(i+n) in the historical output data collected on the similar days to train the first data point of the data to be corrected in the correction model corresponding to the data to be corrected; wherein n is an integer greater than 1;

[0088] 2) Taking n-1 output power data points [p(i+1), p(i+n-1)] corresponding to the data point before the data to be corrected and the predicted output power data point p(i+n) as samples, as well as the meteorological information data of a similar day to the corresponding collection day, predict the second output power data point p(i+n+1) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n+1) in the historical output data collected on the similar day to train the second data point of the data to be corrected in the correction model corresponding to the data to be corrected;

[0089] 3) Repeat the prediction and training process of step 2) above until all output power data points corresponding to the data to be corrected have been predicted and all data points of the data to be corrected in the correction model corresponding to the data to be corrected have been trained, thus completing the training of the correction model corresponding to the data to be corrected.

[0090] In this embodiment, the input layer node of the correction model corresponding to the data to be corrected uses 20 input variables, mainly covering 16 power data sequence points (that is, n=16), 4 meteorological information data, wherein the 4 meteorological information data x(1), x(2), x(3), x(4) are weather type, maximum temperature, minimum temperature and average temperature respectively, and the power data at night are all 0; for local data defects, the 1st to 16th output power data points corresponding to the data points before the data to be corrected (that is, the local defect data) are selected from the historical output data collected on similar days of the corresponding collection day in similar power stations. For example, the data to be corrected is the historical output data of photovoltaic power station A at 6:00-7:00 am on collection day x, and it is determined that the similar power station of photovoltaic power station A is photovoltaic power station B, and the collection day y of photovoltaic power station B is a similar day to collection day x, then the selected 16 output power data points are PV power station B counts from the 1st output power data point to the 16th output power data point at 6:00 a.m. on the collection day y, trains the 17th output power data point, and then trains the 18th data point in combination with the 2nd to 17th output power data points; that is, [p(i), p(i+15)] is used to predict p(i+16), and the 17th output power data point is trained according to the actual value of p(i+16) on the collection day y of PV power station B, and then the new 16 data constructed by [p(i+1), p(i+16)] are used to predict p(i+17), and the 18th output power data point is trained, and so on, until all defective data are predicted and the training of the correction model corresponding to the data to be corrected is completed; among them, the p(i+16) used when predicting p(i+17) is the p(i+16) predicted by [p(i), p(i+15)].

[0091] Since the interference of different equipment models, operating modes and operating environments on the actual output of the power station is eliminated, when the weather conditions are basically similar, such as meteorological days and temperatures, the actual outputs of different photovoltaic power stations also have a certain correlation. Therefore, samples with "similar meteorological days" or "similar weather conditions" in historical information are used, and their output data are used to repair the bad data of the photovoltaic power station; and the output data of photovoltaic power stations located in the same orientation, terrain or geographical location that are particularly close to each other usually have basically the same meteorological conditions, that is, several adjacent power stations in the same geographical area are in extremely similar meteorological conditions, whether it is weather type, temperature, and solar radiation intensity. Therefore, this embodiment uses the data of similar days of the power station for correction, and also uses the historical output data of similar power stations in the same area to assist in the correction of the power station, which can greatly improve the correction accuracy, thereby improving the credibility of the data correction.

[0092] After the correction model corresponding to the data to be corrected is trained, the normal historical output data and meteorological information data corresponding to the day the data to be corrected is collected are input into the correction model to obtain the corrected historical output data; in this embodiment, a neural network is used to construct the correction model. After the neural network is trained, the known power information points and meteorological information data of the day to be repaired are used as input and imported into the trained repair model to obtain the output of power data points at 15-minute intervals, and finally the corrected power output curve is obtained, that is, the corrected historical output data. Figure 2 , Figure 3 Shown is a schematic diagram of the comparison between the historical output data before and after correction and the actual data. It can be seen that the deviation between the corrected historical output data and the actual data is more stable, and there will basically be no sudden large deviation value. In most cases, the deviation between the corrected historical output data and the actual data is smaller than the deviation between the historical output data and the actual data before correction.

[0093] Step 3: Input the corrected historical output data and the pre-processed atmospheric temperature data and solar irradiation data into the prediction model to predict the photovoltaic power generation and obtain the prediction result.

[0094] In this embodiment, the corrected data in step 2 is used in different prediction models for power generation prediction, and the prediction model corresponding to the optimal prediction value is selected for predicting photovoltaic power generation. After comparison, the long short-term memory (LSTM) network is used to predict the corrected data to obtain the prediction result, wherein the LSTM includes an input layer, an output layer and a hidden layer, and the hidden layer includes an LSTM layer composed of LSTM units with unique memory patterns and a fully connected layer with a relu function as an activation function. The LSTM unit has three gate structures, namely, an input gate, a forget gate and an output gate, and a memory unit. The input gate accepts input information and updates the memory unit state according to conditions. The forget gate determines the discarded information according to specific conditions, and the output gate determines the content to be output according to the input information and the memory unit; at each moment, the LSTM unit accepts the current state and the hidden state of the previous moment through three gates, and at the same time receives the state of the memory unit at the previous moment, obtains the state of the new memory unit by calculation, and finally outputs it through the activation function and the output gate.

[0095] The calculation formulas between the variables are as follows:

[0096] The output value of the input gate is:

[0097] i t =σ(W xi x t +W hi h t-1 +W ci c t-1 +bi )

[0098] The output value of the forget gate is:

[0099] f t =σ(W xf x t +W hf h t-1 +W cf c t-1 +b f )

[0100] The output value of the output gate is:

[0101] o t =σ(W xo x t +W ho h t-1 +W co c t +b o )

[0102] The state of the memory unit at the current moment:

[0103] c(t)=f t c t-1 +i t tanh(W xc x t +W hc h t-1 +b c )

[0104] Output of LSTM cell:

[0105] h t =o t tanh(c t )

[0106] Among them, W xf , W hf , W cf , W xi , W hi , W ci , W xc , W hc , W xo , W ho , W co is the weight matrix; b i 、b c 、b f 、b o is the bias parameter; σ is the activation function.

[0107] The photovoltaic power generation prediction method of this embodiment first obtains the historical output data, atmospheric temperature data, and solar irradiation data of distributed photovoltaic power generation systems at different sites, and pre-processes the sample data. Then, the correction model is trained through the output power data corresponding to similar days of similar power stations, so as to correct the bad data (data to be repaired) of the photovoltaic power station. Finally, the long short-term memory (LSTM) network is used to predict the corrected data to obtain the prediction result, which greatly improves the prediction performance of the prediction model.

[0108] Photovoltaic power generation prediction system embodiment

[0109] This embodiment provides a photovoltaic power generation prediction system, including a processor, wherein the processor is used to execute program instructions to implement the photovoltaic power generation prediction method in the above photovoltaic power generation prediction method embodiment.

[0110] Since the specific working principle and working method of the photovoltaic power generation prediction system have been described in detail in the above photovoltaic power generation prediction method embodiment, they will not be repeated here.

[0111] Photovoltaic power generation data correction method embodiment

[0112] This embodiment provides a photovoltaic power generation data correction method, including the following steps:

[0113] The historical output data of each photovoltaic power station in the set geographical area where the photovoltaic power station corresponding to the data to be corrected is located is obtained; based on the historical output data of each photovoltaic power station, the similarity between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected is obtained, and the similar power stations of the photovoltaic power station corresponding to the data to be corrected are determined according to the similarity; in order to measure the numerical proximity and correctly distinguish the shape trend of the curve, a new similarity statistic S that combines the value proximity and the shape proximity is adopted A,B , wherein the angle cosine calculation method is used in the analysis of shape correlation; specifically, in this embodiment, the method of determining similar power stations is as follows:

[0114] In the historical output data of photovoltaic power station A corresponding to the data to be corrected, the output power of photovoltaic power station A corresponding to the data to be corrected is:

[0115] P A = {P A1 ,P A2 ,...,P An}

[0116] In the historical output data of other photovoltaic power stations B, C and D in the same set geographical area, the output power of other photovoltaic power stations B, C and D are:

[0117] P B = {P B1 ,P B2 ,...,P Bn}

[0118] P C = {P C1 ,P C2 ,...,P Cn}

[0119] P D = {P D1 ,P D2 ,...,P Dn}

[0120] The similarity S between the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area is A,B for:

[0121] S A,B =λd A,B +γr A,B

[0122] Where λ and γ are d A,B and r A,B The weight coefficient of ; the appropriate weight coefficient can screen out samples with more accurate similarity, thereby improving the restoration accuracy. In this embodiment, λ and γ are both set to 0.5;

[0123] The photovoltaic power station with a similarity greater than the set similarity threshold and located in the same set geographical area as the photovoltaic power station A corresponding to the data to be corrected is regarded as a similar power station, or the photovoltaic power station with the greatest similarity to the photovoltaic power station A corresponding to the data to be corrected is regarded as a similar power station. A,B and r A,B The calculation formula is as follows:

[0124]

[0125]

[0126] Where n is the number of indicators of the output power sample, k represents the kth indicator, and p ak and p bk are the kth points of the power output of the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area; in this embodiment, the cosine of the angle r is used A,B As a measure of shape similarity, the larger the value, the closer the output power curves of the two PV power stations are to each other.

[0127] Obtain meteorological information data corresponding to a collection day of the data to be corrected, and after determining similar power stations, determine a similar day in the similar power stations that corresponds to the collection day; in this embodiment, the similarity between the meteorological information data of a similar day in the similar power stations that corresponds to the collection day and the meteorological information data corresponding to the collection day is greater than a set threshold; in other embodiments, a collection day in a similar power station that has the same weather type as the collection day of the data to be corrected may also be used as a similar day.

[0128] After determining the similar day corresponding to the collection day in the similar power station, the correction model corresponding to the data to be corrected is trained according to the historical output data collected on the similar day corresponding to the collection day in the similar power station and the meteorological information data on the similar day corresponding to the collection day (in this embodiment, the data to be corrected mainly refers to the data to be corrected corresponding to the local data defects). The specific training process is as follows:

[0129] 1) From the historical output data collected on similar days of the corresponding collection day in similar power stations, select n output power data points [p(i), p(i+n-1)] corresponding to the data point before the data to be corrected as samples, combine the meteorological information data of the similar days of the corresponding collection day, predict the first output power data point p(i+n) corresponding to the data point of the data to be corrected in the historical output data collected on these similar days, and combine the actual value of the output power data point p(i+n) in the historical output data collected on the similar days to train the first data point of the data to be corrected in the correction model corresponding to the data to be corrected; wherein n is an integer greater than 1;

[0130] 2) Taking n-1 output power data points [p(i+1), p(i+n-1)] corresponding to the data point before the data to be corrected and the predicted output power data point p(i+n) as samples, as well as the meteorological information data of a similar day to the corresponding collection day, predict the second output power data point p(i+n+1) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n+1) in the historical output data collected on the similar day to train the second data point of the data to be corrected in the correction model corresponding to the data to be corrected;

[0131] 3) Repeat the prediction and training process of step 2) above until all output power data points corresponding to the data to be corrected have been predicted and all data points of the data to be corrected in the correction model corresponding to the data to be corrected have been trained, thus completing the training of the correction model corresponding to the data to be corrected.

[0132] In this embodiment, the input layer node of the correction model corresponding to the data to be corrected uses 20 input variables, mainly covering 16 power data sequence points (that is, n=16), 4 meteorological information data, wherein the 4 meteorological information data x(1), x(2), x(3), x(4) are weather type, maximum temperature, minimum temperature and average temperature respectively, and the power data at night are all 0; for local data defects, the 1st to 16th output power data points corresponding to the data points before the data to be corrected (that is, the local defective data) are selected from the historical output data collected on similar days of the corresponding collection day in similar power stations. For example, the data to be corrected is the historical output data of photovoltaic power station A from 6:00 to 7:00 a.m. on the collection day x, and it is determined that the similar power station of photovoltaic power station A is photovoltaic power station B, and the collection day of photovoltaic power station B is If day y is a similar day to collection day x, the 16 output power data points selected are the 1st to 16th output power data points of photovoltaic power station B at 6:00 a.m. on collection day y. The 17th output power data point is trained, and then the 18th data point is trained in combination with the 2nd to 17th output power data points; that is, [p(i), p(i+15)] is used to predict p(i+16), and the 17th output power data point is trained according to the actual value of p(i+16) on collection day y of photovoltaic power station B. Then, the new 16 data constructed by [p(i+1), p(i+16)] are used to predict p(i+17), and the 18th output power data point is trained, and so on, until all defective data are predicted and the training of the correction model corresponding to the data to be corrected is completed.

[0133] Since the interference of different equipment models, operating modes and operating environments on the actual output of the power station is eliminated, when the weather conditions are basically similar, such as meteorological days and temperatures, the actual outputs of different photovoltaic power stations also have a certain correlation. Therefore, samples with "similar meteorological days" or "similar weather conditions" in historical information are used, and their output data are used to repair the bad data of the photovoltaic power station; and the output data of photovoltaic power stations located in the same orientation, terrain or geographical location that are particularly close to each other usually have basically the same meteorological conditions, that is, several adjacent power stations in the same geographical area are in extremely similar meteorological conditions, whether it is weather type, temperature, and solar radiation intensity. Therefore, this embodiment uses the data of similar days of the power station for correction, and also uses the historical output data of similar power stations in the same area to assist in the correction of the power station, which can greatly improve the correction accuracy, thereby improving the credibility of the data correction.

[0134] After the correction model corresponding to the data to be corrected is trained, the normal historical output data and meteorological information data corresponding to the day of collection of the data to be corrected are input into the correction model to obtain the corrected historical output data; in this embodiment, a neural network is used to construct a correction model. After the neural network is trained, the known power information points and meteorological information data of the day to be repaired are used as input and imported into the trained repair model to obtain output power data points at intervals of 15 minutes, and finally the corrected power output curve is obtained, that is, the corrected historical output data. Since the correction model contains two data correction references of similar power stations and similar days at the same time, it can constrain the correction of the data to be corrected from two dimensions, thereby improving the correction effect of the correction model and making the corrected historical output data closer to the actual situation.

[0135] Photovoltaic power generation data correction system embodiment

[0136] This embodiment provides a photovoltaic power generation data correction system, including a processor, wherein the processor is used to execute program instructions to implement the photovoltaic power generation data correction method in the above photovoltaic power generation data correction method embodiment.

[0137] Since the specific working principle and working method of the photovoltaic power generation data correction system have been described in detail in the above photovoltaic power generation data correction method embodiment, they will not be repeated here.

[0138] The present invention has the following characteristics:

[0139] 1) On the basis of using the data of similar days of the power station for correction, the historical output data of similar power stations in the same area are also used to assist the power station in correction. That is, the correction model contains both similar power stations and similar days data correction references. Therefore, the correction of the data to be corrected can be constrained from two dimensions, which improves the correction effect of the correction model, makes the corrected historical output data closer to the actual situation, and can greatly improve the correction accuracy, thereby improving the credibility of the data correction.

[0140] 2) A similarity statistic that combines the similarity of numerical values ​​and the similarity of curve shapes is used to judge similar power stations. This can not only measure the similarity of numerical values, but also correctly distinguish the shape trend of the curve.

[0141] It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, but do not constitute a limitation to the present invention.

Claims

1. A photovoltaic power generation data correction method, characterized in that: The steps include: Obtain historical output data of each photovoltaic power station in a set geographical area where the photovoltaic power station corresponding to the data to be corrected is located; obtain similarities between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected, and determine similar power stations to the photovoltaic power station corresponding to the data to be corrected according to the similarities; Obtain meteorological information data corresponding to the collection day of the data to be corrected, and determine a similar day corresponding to the collection day in similar power stations; The similarity between the meteorological information data of a similar day corresponding to the collection day in a similar power station and the meteorological information data corresponding to the collection day is greater than a set threshold; According to the historical output data collected on the similar day of the corresponding collection day in similar power stations and the meteorological information data on the similar day of the corresponding collection day, the correction model corresponding to the data to be corrected is trained; After the training of the correction model corresponding to the data to be corrected is completed, the normal historical output data and meteorological information data corresponding to the collection day of the data to be corrected are input into the correction model to obtain the corrected photovoltaic power generation data.

2. The photovoltaic power generation data correction method according to claim 1, characterized in that: The similarities between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected are obtained, and a method of determining similar power stations of the photovoltaic power station corresponding to the data to be corrected according to the similarities is as follows: In the historical output data of photovoltaic power station A corresponding to the data to be corrected, the output power of photovoltaic power station A corresponding to the data to be corrected is: P A ={P A1 ,P A2 ,...,P An } In the historical output data of other photovoltaic power stations in the same set geographical area, the output power of other photovoltaic power stations B, C and D are: P B ={P B1 ,P B2 ,...,P Bn } P C ={P C1 ,P C2 ,...,P Cn } P D ={P D1 ,P D2 ,...,P Dn } The similarity S between the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area is A,B for: S A,B =λd A,B +γr A,B Where λ and γ are d A,B and r A,B The weight coefficient is: the photovoltaic power station whose similarity is greater than the set similarity threshold and is located in the same set geographical area as the photovoltaic power station A corresponding to the data to be corrected as a similar power station, or the photovoltaic power station A corresponding to the data to be corrected is located in the same set geographical area and has the greatest similarity to the photovoltaic power station A corresponding to the data to be corrected as a similar power station, d A,B and r A,B The calculation formula is as follows: Where n is the number of indicators of the output power sample, k represents the kth indicator, and p ak and p bk They are respectively the k-th point of power output of the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area.

3. The photovoltaic power generation data correction method according to claim 1 or 2, characterized in that: According to the historical output data collected on similar days of the corresponding collection day in similar power stations and the meteorological information data on similar days of the corresponding collection day, the method of training the correction model corresponding to the correction data is as follows: 1) From the historical output data collected on a similar day of the corresponding collection day in similar power stations, select n output power data points [p(i), p(i+n-1)] corresponding to the data point before the data to be corrected as samples, combine the meteorological information data of the similar day of the corresponding collection day, predict the first output power data point p(i+n) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n) in the historical output data collected on the similar day to train the first data point of the data to be corrected in the correction model corresponding to the data to be corrected; wherein n is an integer greater than 1; 2) Taking n-1 output power data points [p(i+1), p(i+n-1)] corresponding to the data point before the data to be corrected and the predicted output power data point p(i+n) as samples, as well as the meteorological information data of a similar day to the corresponding collection day, predict the second output power data point p(i+n+1) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n+1) in the historical output data collected on the similar day to train the second data point of the data to be corrected in the correction model corresponding to the data to be corrected; 3) Repeat the prediction and training process of step 2) above until all output power data points corresponding to the data to be corrected have been predicted and all data points of the data to be corrected in the correction model corresponding to the data to be corrected have been trained, thus completing the training of the correction model corresponding to the data to be corrected.

4. The photovoltaic power generation data correction method according to claim 1 or 2, characterized in that: The meteorological information data includes weather type, maximum temperature, minimum temperature and average temperature.

5. A photovoltaic power generation data correction system, characterized in that: A processor is included, and the processor is used to execute program instructions to implement the photovoltaic power generation data correction method as described in any one of claims 1 to 4 above.

6. A photovoltaic power generation prediction method, characterized in that: The steps include: Acquire and pre-process photovoltaic power generation data, wherein the photovoltaic power generation data includes historical output data of each photovoltaic power station, atmospheric temperature data, and solar irradiation data; Obtaining historical output data of each photovoltaic power station in a set geographical area where the photovoltaic power station corresponding to the preprocessed data to be corrected is located; obtaining similarities between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected, and determining similar power stations to the photovoltaic power station corresponding to the data to be corrected according to the similarities; Obtain meteorological information data corresponding to the collection day of the data to be corrected, and determine a similar day corresponding to the collection day in similar power stations; The similarity between the meteorological information data of a similar day corresponding to the collection day in a similar power station and the meteorological information data corresponding to the collection day is greater than a set threshold; According to the historical output data collected on the similar day of the corresponding collection day in similar power stations and the meteorological information data on the similar day of the corresponding collection day, the correction model corresponding to the data to be corrected is trained; After the correction model training corresponding to the data to be corrected is completed, the normal historical output data and meteorological information data corresponding to the collection day of the data to be corrected are input into the correction model to obtain the corrected historical output data; The corrected historical output data and the pre-processed atmospheric temperature data and solar irradiation data are input into the prediction model to predict the photovoltaic power generation and obtain the prediction results.

7. The photovoltaic power generation prediction method according to claim 6, characterized in that: The similarities between other photovoltaic power stations in the same set geographical area as the photovoltaic power station corresponding to the data to be corrected and the photovoltaic power station corresponding to the data to be corrected are obtained, and a method of determining similar power stations of the photovoltaic power station corresponding to the data to be corrected according to the similarities is as follows: In the historical output data of photovoltaic power station A corresponding to the data to be corrected, the output power of photovoltaic power station A corresponding to the data to be corrected is: P A ={P A1 ,P A2 ,...,P An } In the historical output data of other photovoltaic power stations in the same set geographical area, the output power of other photovoltaic power stations B, C and D are: P B ={P B1 ,P B2 ,...,P Bn } P C ={P C1 ,P C2 ,...,P Cn } P D ={P D1 ,P D2 ,...,P Dn } The similarity S between the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B in the same set geographical area is A,B for: S A,B =λd A,B +γr A,B Where λ and γ are d A,B and r A,B The weight coefficient is: the photovoltaic power station whose similarity is greater than the set similarity threshold and is located in the same set geographical area as the photovoltaic power station A corresponding to the data to be corrected as a similar power station, or the photovoltaic power station A corresponding to the data to be corrected is located in the same set geographical area and has the greatest similarity to the photovoltaic power station A corresponding to the data to be corrected as a similar power station, d A,B and r A,B The calculation formula is as follows: Where n is the number of indicators of the output power sample, k represents the kth indicator, and p ak and p bk They are respectively the k-th point of power output of the photovoltaic power station A corresponding to the data to be corrected and the photovoltaic power station B located in the same set geographical area.

8. The photovoltaic power generation prediction method according to claim 6 or 7, characterized in that: According to the historical output data collected on similar days of the corresponding collection day in similar power stations and the meteorological information data on similar days of the corresponding collection day, the method of training the correction model corresponding to the correction data is as follows: 1) From the historical output data collected on a similar day of the corresponding collection day in similar power stations, select n output power data points [p(i), p(i+n-1)] corresponding to the data point before the data to be corrected as samples, combine the meteorological information data of the similar day of the corresponding collection day, predict the first output power data point p(i+n) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n) in the historical output data collected on the similar day to train the first data point of the data to be corrected in the correction model corresponding to the data to be corrected; wherein n is an integer greater than 1; 2) Taking n-1 output power data points [p(i+1), p(i+n-1)] corresponding to the data point before the data to be corrected and the predicted output power data point p(i+n) as samples, as well as the meteorological information data of a similar day to the corresponding collection day, predict the second output power data point p(i+n+1) corresponding to the data point of the data to be corrected in the historical output data collected on the similar day, and combine the actual value of the output power data point p(i+n+1) in the historical output data collected on the similar day to train the second data point of the data to be corrected in the correction model corresponding to the data to be corrected; 3) Repeat the prediction and training process of step 2) above until all output power data points corresponding to the data to be corrected have been predicted and all data points of the data to be corrected in the correction model corresponding to the data to be corrected have been trained, thus completing the training of the correction model corresponding to the data to be corrected.

9. The photovoltaic power generation prediction method according to claim 6 or 7, characterized in that: The meteorological information data includes weather type, maximum temperature, minimum temperature and average temperature.

10. A photovoltaic power generation prediction system, characterized in that: It comprises a processor, wherein the processor is used to execute program instructions to implement the photovoltaic power generation prediction method as described in any one of claims 6 to 9.

Images