A Method for Predicting Line Losses in a Distributed Power Station

Abstract

The present invention provides a method for predicting the line loss of a distributed power station, which includes the following steps: collecting the time-series output voltage, wind direction, wind speed, and light intensity data of each photovoltaic field; classifying the collected data; forming the grouped data into sample data and training a neural network model to predict the change rate of the voltage in the next cycle of the photovoltaic field to be measured; calculating the output voltage in the next cycle, and predicting and calculating the line loss in the next cycle of the photovoltaic field to be measured based on the output voltage in the next cycle. Through the above solution, the technical problem of inaccurate line loss prediction calculation is solved.

Description

Technical Field

[0001] The present invention relates to the field of transmission line loss calculation, and particularly to a line loss prediction calculation method and system based on machine learning. Background Art

[0002] Although good conductors such as copper and aluminum are used as conductors for power transmission in transmission lines, they still have a certain resistance value. Especially in the case of relatively long lines, the accumulated resistance value is also quite considerable, and the consumed electric energy can account for seven or eight percent of the transmitted electric energy. According to the principle of power consumption, when the transmission line is fixed, that is, when the resistance is constant, if the voltage level is low, the loss is higher. Distributed photovoltaic power stations are usually distributed over a large area, and there may be a long distance between the photovoltaic field and the inverter. Therefore, the line resistance is large. At the same time, the output voltage of the photovoltaic field is generally low, so the line loss will be relatively large. Due to the large fluctuations in the photovoltaic power grid, in order to output stable power, power compensation is usually required. In order to accurately control the power of the power grid, it is very important to accurately predict and calculate the change of line loss.

[0003] However, photovoltaic power generation is affected by meteorological factors such as irradiance, sunshine duration, and cloud cover. When the light fluctuates, the output voltage of the photovoltaic field will fluctuate. Therefore, the line loss is not fixed and there is an ultra-short-term fluctuation problem. According to the loss principle, predicting the output voltage is the key to accurately calculating the line loss. At present, the ultra-short-term prediction of the photovoltaic field is mainly carried out through historical weather, real-time weather, historical sunshine, real-time sunshine, etc. For example, CN102281016A discloses a clear sky photovoltaic ultra-short-term power prediction method based on real-time radiation acquisition technology, and CN101969207A discloses a photovoltaic ultra-short-term power prediction method combining satellite remote sensing and meteorological telemetry technology. The above methods mainly predict based on the natural conditions affecting photovoltaic power generation, but the natural conditions themselves have a large randomness. Although weather forecasts are becoming more and more accurate, they are still one of the most difficult problems in the world. Simply using natural conditions for prediction will inevitably result in large errors, causing inaccurate calculation of line loss. Summary of the Invention

[0004] In order to solve the problem of large ultra-short-term line loss calculation error in the photovoltaic system, the present invention proposes a line loss prediction method for distributed power stations.

[0005] In one aspect of the present invention, a method for predicting line loss in a distributed power station is proposed. The line loss prediction calculation method is applied to a distributed photovoltaic field, and the distributed photovoltaic field includes multiple independent photovoltaic fields. It is characterized in that the line loss prediction calculation method includes the following steps: Synchronously collect the output voltage, wind direction, wind speed, and light intensity of each independent photovoltaic field in the first period to obtain the time-series output voltage, wind direction, wind speed, and light intensity data of each independent photovoltaic field; Calculate the change rate sequence for the time-series output voltage data of each independent photovoltaic field, classify the change rate, and classify the wind direction, wind speed, and light intensity to obtain a classification sequence; Use the classification of the output voltage change in the (T + 1)-th period of the photovoltaic field to be measured as a label, and use the output voltage change, wind direction, wind speed, light intensity classification sequences in the T-th period of all other photovoltaic fields and the coordinates of the corresponding photovoltaic fields as input to form sample data, and use the sample data to train a neural network model to obtain the trained neural network model of the photovoltaic field to be measured; where T is the number of the classification sequence, taking values of 1, 2, 3, 4... N - 1, and N is the total data volume; Real-time collect the output voltage change, wind direction, wind speed, and light intensity data of each independent photovoltaic field and process them into classified data, input the real-time classified data of all photovoltaic fields except the photovoltaic field to be measured and the corresponding photovoltaic field coordinate data into the trained neural network model of the photovoltaic field to be measured, predict the classification of the output voltage change in the next period of the photovoltaic field to be measured, and then determine the change rate of the voltage in the next period; Calculate the output voltage in the next period according to the real-time output voltage of the photovoltaic field to be measured and the change rate of the voltage in the next period, and predict the line loss in the next period of the photovoltaic field to be measured based on the output voltage in the next period.

[0006] Further, a distributed photovoltaic field refers to having more than 10 independent photovoltaic sites within the same photovoltaic management system. The same photovoltaic management system means that each photovoltaic field is connected to the same transmission and dispatching system and is managed by the same system.

[0007] Further, remove the sample data with all output voltage change rates being 0.

[0008] Further, each independent photovoltaic field in the distributed photovoltaic field is sequentially set as the photovoltaic field to be measured, and a model is trained for each independent photovoltaic field.

[0009] Further, one data is collected every 2 minutes starting from sunrise every day, and data for a whole year is collected.

[0010] Another method of the present invention proposes a line loss prediction calculation system. The line loss prediction calculation system is applied to a distributed photovoltaic field, and the distributed photovoltaic field includes multiple independent photovoltaic fields. It is characterized in that:

[0011] The line loss prediction calculation system includes, for example: an acquisition module that synchronously acquires the output voltage, wind direction, wind speed, and light intensity of each independent photovoltaic field at a first period to obtain time - serialized output voltage, wind direction, wind speed, and light intensity data of each independent photovoltaic field; a grading module that calculates the change rate sequence of the time - serialized output voltage data of each independent photovoltaic field, grades the change rate, and grades the wind direction, wind speed, and light intensity to obtain a grading sequence; a training module that uses the grading of the output voltage change in the (T + 1) - th period of the photovoltaic field to be measured as a label, and uses the grading sequences of the output voltage changes, wind direction, wind speed, and light intensity in the T - th period of all other photovoltaic fields and the coordinates of the corresponding photovoltaic fields as input to form sample data, and uses the sample data to train a neural network model to obtain the trained neural network model of the photovoltaic field to be measured; where T is the number of the grading sequence, taking values of 1, 2, 3, 4... N - 1, and N is the total data volume; a prediction module that real - time acquires the output voltage changes, wind direction, wind speed, and light intensity data of each independent photovoltaic field and processes them into graded data, inputs the real - time graded data of the photovoltaic field to be measured and the corresponding photovoltaic field coordinate data into the trained neural network model of the photovoltaic field to be measured, predicts the grading of the output voltage change in the next period of the photovoltaic field to be measured, and further determines the change rate of the voltage in the next period; a calculation module that calculates the output voltage in the next period according to the real - time output voltage of the photovoltaic field to be measured and the change rate of the voltage in the next period, and predicts and calculates the line loss in the next period of the photovoltaic field to be measured based on the output voltage in the next period.

[0012] Furthermore, a distributed photovoltaic field refers to a situation where there are more than 10 independent photovoltaic sites within the same photovoltaic management system. The same photovoltaic management system means that each photovoltaic field is connected to the same transmission and dispatching system and is managed by the same system.

[0013] Furthermore, sample data with an output voltage change rate of all 0 is removed.

[0014] Furthermore, each independent photovoltaic field in the distributed photovoltaic field is sequentially set as the photovoltaic field to be measured, and a model is trained for each independent photovoltaic field.

[0015] Furthermore, one data is collected every 2 minutes starting from sunrise every day, and data for a whole year is collected.

[0016] Through the above - mentioned technical solution, the present invention predicts the ultra - short - term output voltage change of the photovoltaic field to be measured according to the real - time output voltage change rate of other photovoltaic fields in the distributed photovoltaic field and real - time conditions such as sunlight and wind direction, avoids the inaccurate problem caused by directly using meteorological data, and improves the accuracy of predicting and calculating the line loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 Schematic diagram of two photovoltaic fields. DETAILED DESCRIPTION

[0019] Below, the invention is preferably described in conjunction with the accompanying drawings and specific implementation methods.

[0020] like Figure 1 As shown, the output voltage of photovoltaic field A decreases due to the cover of dark clouds, and photovoltaic field B is not affected at this time. However, as time goes by, the dark clouds will move to photovoltaic field B, and will have a similar impact on photovoltaic field B in a certain period of time thereafter. Therefore, the output voltage change rate of photovoltaic field B can be predicted based on the location, wind direction, light, and output voltage change rate of photovoltaic field A. Based on this principle, the present application predicts the subsequent output voltage change of photovoltaic field B through the current output voltage change of photovoltaic field A. For a system with many photovoltaic sites, as long as the output of one site changes, the changes of other scenes in the subsequent time can be predicted in time, so that the system can make timely adjustments. The specific implementation method is as follows:

[0021] Embodiment 1 provides a distributed power station line loss prediction method, which is applied to a distributed photovoltaic field, wherein the distributed photovoltaic field includes a plurality of independent photovoltaic fields.

[0022] Distributed photovoltaic fields are relative to centralized photovoltaic fields. Centralized large-area photovoltaics are built in areas such as deserts and Gobi, and photovoltaic fields are built in blocks, while distributed photovoltaic fields make full use of idle land resources and may be distributed on roofs, water surfaces, etc.; distributed photovoltaic fields include multiple unconnected and independent photovoltaic electric fields, and the size, output power, location coordinates, etc. of each photovoltaic electric field may be different. For example, it may include several rooftop electric fields of ordinary users, several electric fields on the surface of fishery aquaculture, etc.; the output voltage of distributed photovoltaic fields is relatively low, so the line loss fluctuates greatly; further, the distributed photovoltaic field described in this application refers to a photovoltaic management system with more than 10 independent photovoltaic sites. The same photovoltaic management system is that each photovoltaic field is connected to the same transmission scheduling system and is managed by the same system; distributed photovoltaic fields are usually characterized by a wide range and uneven distribution; distributed photovoltaic fields managed by the same photovoltaic management center may be distributed over hundreds of square kilometers, so the meteorological conditions between distributed photovoltaic fields may be very different. At the same time, some photovoltaic fields may be covered by dark clouds, while others may be sunny.

[0023] Output voltage, wind direction, wind speed, and light intensity are collected for each of the independent photovoltaic fields synchronously in the first period to obtain time - serialized output voltage, wind direction, wind speed, and light intensity data for each independent photovoltaic field;

[0024] For the convenience of subsequent data processing, the output voltage of each independent photovoltaic field is collected synchronously at the same period.

[0025] Exemplarily, voltage data is collected with a 2 - minute time period, that is, the output voltage of the photovoltaic scenario is collected once every 2 minutes.

[0026] Exemplarily, for the purpose of maintaining synchronization, all photovoltaic fields adopt the following time series (with a 2 - minute period): 12:00, 12:02, 12:04, …… 13:02, 13:04, …… 15:22, 15:24 …… for data collection; In the example, starting from 12 o'clock, the voltage is collected every 2 minutes.

[0027] The first period can be freely set. Obviously, the shorter the period, the denser the data, and the higher the prediction accuracy using the present invention, but the training of the model will be slower; the longer the period, the less data, and although a certain accuracy is sacrificed, the processing speed will be greatly improved. Those skilled in the art can freely choose according to the system hardware conditions and the requirements for accuracy, and the present application does not make excessive limitations.

[0028] Based on the aforementioned time series, exemplarily, photovoltaic field A obtains a series of voltage data: 376, 370, 365, ……, which correspond to the data collected at time points 12:00, 12:02, 12:04 ……

[0029] Similarly, data is collected for all independent photovoltaic fields in the management system to obtain a series of time - series data of the output voltage for each independent photovoltaic field.

[0030] Similar to the method of collecting the output voltage, the wind direction, wind speed, and light intensity of each independent photovoltaic field are collected at the same first period. For the purpose of maintaining time synchronization, similarly, data is collected at the same time points and time intervals.

[0031] Calculate the change - rate sequence for the time - serialized output voltage data of each independent photovoltaic field, classify the change rate, and classify the wind direction, wind speed, and light intensity to obtain a classification sequence.

[0032] On the one hand, due to the influence of weather changes, the absolute change amounts between different photovoltaic sites may vary greatly, but the change ratios are similar. On the other hand, for the transmission management system, given the output voltages of all current photovoltaic sites, when making prediction adjustments, it only needs to care about the voltage after one adjustment cycle, that is, the management system is more concerned about the short-term future voltage fluctuations so as to take countermeasures in advance. Therefore, the management system is more concerned about the voltage change rate. At the same time, the absolute values of the output voltages of different sites vary greatly. If the absolute values of the voltages are directly used for prediction training, the dimensionality of the training data will increase exponentially, resulting in an extremely long training process. However, the change rate is usually between one and two percentage points, with a limited change range, making it easier to set levels. After classification, the overall data volume will be greatly reduced.

[0033] Therefore, to accelerate the training speed, further, a change rate sequence is calculated based on the output voltage of each photovoltaic field, and the absolute value of the voltage is filtered out by the change rate.

[0034] Exemplarily, photovoltaic field A obtains a series of data: 376, 370, 365, 369, ……, and the result of the change rate sequence is 0, -0.01596, -0.01351, 0.01096 …….

[0035] Further, to facilitate data training, the change rate is classified into change levels; exemplarily, the change rate is classified as described in Table 1.

[0036] Table 1

[0037]

[0038]

[0039] Then the classified sequence after classifying the above change rate is 0, -4, -3, 3 …….

[0040] All photovoltaic fields are processed with the above classified sequence to obtain the classified sequence of each independent photovoltaic field. The classified sequence is expressed, unifying the voltage changes to the same dimension, greatly reducing the training calculation amount and improving the training speed.

[0041] Preferably, similar to the processing of voltage data, to reduce the data volume and accelerate the training speed, the wind direction, wind speed, and light intensity are classified to obtain corresponding grouped sequences.

[0042] Exemplarily, the wind direction is divided into eight directions: north, northeast, east, southeast, south, southwest, west, and northwest.

[0043] Exemplarily, the wind speed is classified every 0.5 level, such as 0, 0.5, 1, 1.5, 2, 2.5 …….

[0044] Exemplarily, the light intensity is graded every 0.5×10⁴ lux, for example, 0 - 0.5×10⁴ lux is level 1, 0.5×10⁴ to 1×10⁴ lux is level 2, 1×10⁴ to 1.5×10⁴ lux is level 3...

[0045] Taking the output voltage change grading of the (T + 1)-th cycle of the photovoltaic field to be measured as the label, and taking the output voltage changes, wind directions, wind speeds, light intensity grading sequences of the T-th cycles of all other photovoltaic fields and the coordinates of the corresponding photovoltaic fields as inputs to form sample data, and using the sample data to train a neural network model to obtain the trained neural network model of the photovoltaic field to be measured; where T is the number of the grading sequence, taking values of 1, 2, 3, 4... N - 1, and N is the total amount of data.

[0046] Exemplarily, the output voltage change of the (T + 1)-th cycle of photovoltaic field A relative to the T-th cycle is level 3. More specifically, for example, at 13:02 in the (T + 1)-th cycle in photovoltaic field A, the output voltage change relative to 13:00 in the T-th cycle is level 3; in the T-th cycle, that is, at 13:00, the output voltage changes, wind directions, wind speeds, light intensities, and coordinates of several other photovoltaic fields are (2, east, 1, 9, (73.2, 94.3)), (2, east, 1, 10, (73.2, 93.3)), (1, east, 1, 10, (73.7, 95.3)), (1, northeast, 1, 9, (76.2, 92.4))...

[0047] Then, as shown in the following formula, the output voltage changes, wind directions, wind speeds, light intensities, and coordinates of the T-th cycles of several other photovoltaic fields and the voltage change of the T-th cycle of the current photovoltaic field to be measured form a sample

[0048] Label - level 3

[0049] All the training data can be obtained by the same method, and after obtaining enough training data, model training can be carried out.

[0050] Preferably, in order to cover all seasonal variations, at least one year's data is obtained as training data; specifically, in one year, one data can be collected every 2 minutes starting from sunrise every day, and data for a whole year is collected, and the total number of collections is the total amount of data N.

[0051] Preferably, in order to reduce the amount of data, the sample data with all output voltage change rates being 0 is removed, that is, when all the output voltage changes in a sample are 0, which means the output voltage does not change at the sampling moment, such data has no change meaning and cannot play a role in prediction, so it is removed.

[0052] Train using sample data to obtain a trained neural network model. For a specific neural network model, the present application does not make specific limitations. Since the data has been processed, most neural networks in the prior art can be applied to the present application, such as CNN, GAN networks, etc.

[0053] Furthermore, each independent photovoltaic field in the distributed photovoltaic field is sequentially set as a photovoltaic field to be measured, and a model is trained for each independent photovoltaic field, so that each independent photovoltaic field can be predicted.

[0054] Real-time collect the output voltage change, wind direction, wind speed, and light intensity data of each independent photovoltaic field, and perform preprocessing. Input the preprocessed output voltage change, wind direction, wind speed, light intensity, and coordinate data outside the photovoltaic field to be measured into the trained neural network model to predict the output voltage change level of the photovoltaic field to be measured in the next cycle, and then determine the change rate of the voltage.

[0055] Exemplarily, real-time collect the output voltage change, wind direction, wind speed, and light intensity data of each independent photovoltaic field, and perform the same processing as the training data, that is, classify the voltage change, wind direction, wind speed, and light intensity; a more specific example is to use the data of other photovoltaic fields in the system at 15:00 as (1, west, 1, 9, (73.2, 94.3)), (0, northwest, 1, 10, (73.2, 93.3)), (0, west, 1, 10, (73.7, 95.3)), (1, northwest, 1, 9, (76.2, 92.4))... Input the trained neural network model of the photovoltaic field to be measured, the output change level is level 1, and the output is the prediction result of the next cycle of the photovoltaic field to be predicted, then the predicted change rate of the photovoltaic field to be measured in the next cycle is +0.5%.

[0056] Calculate the output voltage of the next cycle according to the real-time output voltage of the photovoltaic field to be measured and the output voltage change rate of the next cycle, and predict the loss of the photovoltaic field to be measured based on the output voltage prediction of the next cycle.

[0057] Exemplarily, the voltage output by the photovoltaic field to be measured at 15:00 is 376 volts, and the predicted change rate is +0.5%, then the predicted voltage of the next cycle is 376*(1 + 0.5%), and the predicted line loss can be calculated through the predicted voltage.

[0058] Embodiment 2, on the other hand, proposes a line loss prediction calculation system, which is applied to a distributed photovoltaic field, and the distributed photovoltaic field includes multiple independent photovoltaic fields.

[0059] Distributed photovoltaic fields are relative to centralized photovoltaic fields. Centralized large-area photovoltaics are built in areas such as deserts and Gobi, and photovoltaic fields are built in blocks, while distributed photovoltaic fields make full use of idle land resources and may be distributed on roofs, water surfaces, etc.; distributed photovoltaic fields include multiple unconnected and independent photovoltaic electric fields, and the size, output power, location coordinates, etc. of each photovoltaic electric field may be different. For example, it may include several rooftop electric fields of ordinary users, several electric fields on the surface of fishery aquaculture, etc.; the output voltage of distributed photovoltaic fields is relatively low, so the line loss fluctuates greatly; further, the distributed photovoltaic field described in this application refers to a photovoltaic management system with more than 10 independent photovoltaic sites. The same photovoltaic management system is that each photovoltaic field is connected to the same transmission scheduling system and is managed by the same system; distributed photovoltaic fields are usually characterized by a wide range and uneven distribution; distributed photovoltaic fields managed by the same photovoltaic management center may be distributed over hundreds of square kilometers, so the meteorological conditions between distributed photovoltaic fields may be very different. At the same time, some photovoltaic fields may be covered by dark clouds, while others may be sunny.

[0060] The system comprises:

[0061] The acquisition module synchronously acquires the output voltage, wind direction, wind speed, and light intensity of each independent photovoltaic field in a first cycle to obtain the time-series output voltage, wind direction, wind speed, and light intensity data of each independent photovoltaic field;

[0062] In order to facilitate subsequent data processing, the output voltage of each independent photovoltaic field is collected synchronously with the same cycle.

[0063] Exemplarily, the voltage data is collected in a 2-minute period, that is, the output voltage of the photovoltaic scene is collected once every 2 minutes.

[0064] Exemplarily, in order to maintain synchronization, all photovoltaic fields use the following time series (with a period of 2 minutes) 12:00, 12:02, 12:04, ... 13:02, 13:04, ... 15:22, 15:24 ... for data collection; in the example, starting from 12 o'clock, the voltage is collected every 2 minutes.

[0065] The first cycle can be set freely. Obviously, the shorter the cycle is, the denser the data is, and the higher the prediction accuracy of the present invention is, but the training of the model will be slower; the longer the cycle is, the less data is, although a certain degree of accuracy is sacrificed, the processing speed will be greatly improved. Technical personnel in this field can freely choose according to the system hardware conditions and the requirements for accuracy, and this application does not make too many restrictions.

[0066] Based on the aforementioned time series, exemplarily, photovoltaic field A obtained a series of voltage data: 376, 370, 365, ……, which correspond to the data collected at time points 12:00, 12:02, 12:04 ……

[0067] Similarly, data is collected for all independent photovoltaic fields within the management system to obtain a series of output voltage time series data for each independent photovoltaic field.

[0068] Similar to the method of collecting output voltage, the wind direction, wind speed, and light intensity of each independent photovoltaic field are collected with the same first period. To maintain time synchronization, similarly, data is collected at the same time points and time intervals.

[0069] The grading module calculates the change rate sequence for the time - serialized output voltage data of each independent photovoltaic field, grades the change rate, and grades the wind direction, wind speed, and light intensity to obtain a grading sequence.

[0070] On the one hand, due to the influence of weather changes, the absolute change amounts between different photovoltaic sites may vary greatly, but the change ratios are similar; on the other hand, for the transmission management system, given the output voltages of all current photovoltaic sites, when making prediction adjustments, it only needs to care about the voltage after one adjustment period, that is, the management system is more concerned about the short - term future voltage fluctuations in order to take countermeasures in advance. Therefore, the management system is more concerned about the change rate of voltage; at the same time, the absolute values of the output voltages of different sites vary greatly. If the absolute values of voltage are directly used for prediction training, the dimensionality of the training data will increase exponentially, resulting in an extremely long training process; while the change rate is usually within one to two percentage points, with a limited range of change, it is relatively easy to set grades, and the overall data volume after grading will be greatly reduced.

[0071] Therefore, in order to speed up the training speed, further, the change rate sequence is calculated based on the output voltage of each photovoltaic field, and the change rate filters out the absolute value of the voltage.

[0072] Exemplarily, photovoltaic field A obtained a series of data: 376, 370, 365, 369, ……, then the results of the change rate sequence are 0, - 0.01596, - 0.01351, 0.01096 …….

[0073] Further, to facilitate data training, the change rate is classified into change levels; exemplarily, the change rate is classified as described in Table 1.

[0074] Table 1

[0075] Greater than 1.5% 4 +1% to 1.5% 3 +0.5% to +1% 2 0 to +0.5% 1 0 0 -0.5% to 0 -1 -1% to -0.5% -2 -1% to -1.5% -3 Less than -1.5% -4

[0076] Then the grading sequence after grading the above change rate is 0, -4, -3, 3...

[0077] All photovoltaic power stations are processed according to the above grading sequence to obtain the grading sequence of each independent photovoltaic power station. The grading sequence is expressed, unifying the voltage change to the same dimension, greatly reducing the training calculation amount and improving the training speed.

[0078] Preferably, similar to the processing of voltage data, in order to reduce the data volume and speed up the training speed, the wind direction, wind speed, and light intensity are graded to obtain the corresponding grouping sequences.

[0079] Exemplarily, the wind direction is divided into eight directions: north, northeast, east, southeast, south, southwest, west, and northwest.

[0080] Exemplarily, the wind speed is classified every 0.5 level, such as 0, 0.5, 1, 1.5, 2, 2.5...

[0081] Exemplarily, the light intensity is graded every 0.5 million lux, such as 0 - 0.5 million lux is level 1, 0.5 million to 1 million lux is level 2, 1 million to 1.5 million lux is level 3...

[0082] The training module uses the output voltage change grading of the (T + 1)-th cycle of the photovoltaic power station to be measured as the label, and uses the output voltage changes, wind directions, wind speeds, light intensities, and grading sequences of the (T)-th cycle of all other photovoltaic power stations and the coordinates of the corresponding photovoltaic power stations as inputs to form sample data, and uses the sample data to train the neural network model to obtain the trained neural network model of the photovoltaic power station to be measured; where T is the number of the grading sequence, and the value ranges from 1, 2, 3, 4... N - 1, and N is the total data volume.

[0083] Exemplarily, the output voltage change of photovoltaic power station A in the (T + 1)-th cycle relative to the T-th cycle is level 3. More specifically, in photovoltaic power station A, at 13:02 in the (T + 1)-th cycle, the output voltage change relative to 13:00 in the T-th cycle is level 3; in the T-th cycle, that is, at 13:00, the output voltage changes, wind directions, wind speeds, light intensities, and coordinates of several other photovoltaic power stations are (2, east, 1, 9, (73.2, 94.3)), (2, east, 1, 10, (73.2, 93.3)), (1, east, 1, 10, (73.7, 95.3)), (1, northeast, 1, 9, (76.2, 92.4))...

[0084] Then as shown in the following formula, the output voltage changes, wind directions, wind speeds, light intensities, and coordinates of several other photovoltaic power stations in the T-th cycle and the voltage change of the current photovoltaic power station to be measured in the T-th cycle form a sample

[0085] All the training data can be obtained by the same method. After obtaining enough training data, model training can be carried out.

[0086] Preferably, in order to cover all seasonal variations, at least one year's data is obtained as training data. Specifically, within one year, data can be collected every 2 minutes starting from sunrise every day for a whole year, and the total number of collections is the total data volume N.

[0087] Preferably, in order to reduce the data volume, sample data with an output voltage change rate of all 0 is removed. That is, when the output voltage change in a sample is all 0, which means the output voltage does not change at the sampling moment, this kind of data has no change meaning and cannot play a role in prediction, so it is removed.

[0088] Training is carried out using the sample data to obtain a trained neural network model. For the specific neural network model, this application does not make specific limitations. Since the data has been processed, most neural networks in the prior art can be applied to this application, such as CNN, GAN networks, etc.

[0089] Furthermore, each independent photovoltaic field in the distributed photovoltaic field is sequentially set as the photovoltaic field to be measured, and a model is trained for each independent photovoltaic field, so that predictions can be made for each independent photovoltaic field.

[0090] The prediction module collects in real time the output voltage change, wind direction, wind speed, and light intensity data of each independent photovoltaic field, and performs preprocessing. The preprocessed output voltage change, wind direction, wind speed, light intensity, and coordinate data of the photovoltaic field to be measured are input into the trained neural network model to predict the output voltage change level of the photovoltaic field to be measured in the next cycle, and then the voltage change rate is determined.

[0091] Exemplarily, the output voltage change, wind direction, wind speed, and light intensity data of each independent photovoltaic field are collected in real time and processed in the same way as the training data, that is, the voltage change, wind direction, wind speed, and light intensity are classified. A more specific example is to use the data of other photovoltaic fields in the system at 15:00 as (1, west, 1, 9, (73.2, 94.3)), (0, northwest, 1, 10, (73.2, 93.3)), (0, west, 1, 10, (73.7, 95.3)), (1, northwest, 1, 9, (76.2, 92.4))... and input them into the trained neural network model of the photovoltaic field to be measured. The output change level is level 1, and the output is the prediction result of the next cycle of the photovoltaic field to be predicted. Then, the change rate of the photovoltaic field to be measured in the next cycle is predicted to be +0.5%.

[0092] A calculation module calculates the output voltage of the next cycle based on the real-time output voltage of the photovoltaic field to be measured and the output voltage change rate of the next cycle, and predicts the loss of the photovoltaic field to be measured based on the output voltage of the next cycle.

[0093] Exemplarily, the voltage output by the photovoltaic field to be measured at 15:00 is 376 volts, and the predicted change rate is +0.5%. Then the predicted voltage of the next cycle is 376*(1 + 0.5%). The predicted line loss can be calculated through the predicted voltage.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific implementation manners of the present invention or make equivalent replacements. Any modification or equivalent replacement without departing from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

[0095] For the part of the module structure not specifically defined in the present invention, the content recorded in the prior art shall prevail. The prior art mentioned in the foregoing background art part and the specific embodiment part of the present invention can be used as a part of the present invention to understand the meaning of some technical features or parameters. The protection scope of the present invention shall be subject to the content actually recorded in the claims.

Claims

1. A method for predicting line loss in a distributed power station, the line loss prediction calculation method is applied to a distributed photovoltaic field, and the distributed photovoltaic field includes multiple independent photovoltaic fields; characterized in that: The line loss prediction calculation method includes the following steps: S1. Synchronously collect the output voltage, wind direction, wind speed, and light intensity of each independent photovoltaic field in the first period to obtain the time-series output voltage, wind direction, wind speed, and light intensity data of each independent photovoltaic field; S2. Calculate the change rate sequence for the time-series output voltage data of each independent photovoltaic field and remove the sample data with all output voltage change rates being 0. Then, classify the change rate and classify the wind direction, wind speed, and light intensity to obtain the classification sequence; S3. Use the output voltage change classification of the (T + 1)-th period of the photovoltaic field to be measured as the label, and use the output voltage change, wind direction, wind speed, and light intensity classification sequences of the T-th period of all other photovoltaic fields and the coordinates of the corresponding photovoltaic fields as input to form sample data. Use the sample data to train a neural network model to obtain the trained neural network model of the photovoltaic field to be measured; where T is the number of the classification sequence, taking values of 1, 2, 3, 4... N - 1, and N is the total data volume; S4. Real-time collect the output voltage change, wind direction, wind speed, and light intensity data of each independent photovoltaic field and process them into classified data. Input the real-time classified data of the photovoltaic field to be measured and the corresponding photovoltaic field coordinate data into the trained neural network model of the photovoltaic field to be measured to predict the output voltage change classification of the next period of the photovoltaic field to be measured, and then determine the change rate of the voltage in the next period; For each independent photovoltaic field in the distributed photovoltaic field, set it as the photovoltaic field to be measured in turn, and train a model for each independent photovoltaic field; S5. Calculate the output voltage of the next period according to the real-time output voltage of the photovoltaic field to be measured and the change rate of the voltage in the next period, and predict the line loss of the next period of the photovoltaic field to be measured based on the output voltage of the next period.

2. The distributed power station line loss prediction method according to claim 1, characterized in that: A distributed photovoltaic field refers to a photovoltaic field with more than 10 independent photovoltaic fields within the same photovoltaic management system. The same photovoltaic management system means that each photovoltaic field is connected to the same transmission and dispatching system and is managed by the same transmission and dispatching system.

3. The distributed power station line loss prediction method according to claim 1, characterized in that: Collect one data every 2 minutes starting from sunrise every day and collect data for a whole year.

4. The distributed power station line loss prediction method according to claim 1, characterized in that: Classify the wind speed by every 0.5 level.

5. A method for predicting line loss of a distributed power station according to claim 1, characterized in that: Classify the light intensity by every 0.5 million lux.

Images