Early warning method for sudden power change of photovoltaic power station based on data of ground meteorological station

Abstract

The invention discloses a photovoltaic power station power sudden change early warning method based on ground meteorological station data, and belongs to the technical field of solar power generation. The method comprises the steps of data reading and preprocessing, key index calculation, dynamic threshold reading, logic judgment, early warning information generation and release and historical data queue updating. The method comprises the following steps: regularly reading total radiation GHI and scattered radiation DIF data of an existing meteorological station of a photovoltaic power station, calculating a scattering proportion Kt and a total radiation decline rate R after preprocessing, and constructing a dual-condition judgment logic by combining dynamic thresholds Rth and Kth, rth and a scattering proportion change rate dKdt gt; and when Kth is greater than Kth, generating and issuing early warning information. Hardware does not need to be newly added, and the cost is low; the device is not influenced by severe weather and has high reliability; the algorithm is simple, and the response real-time performance is high; the early warning mechanism is clear, the accuracy is high, buffer time can be provided for power grid dispatching and energy storage adjustment, and the friendly grid connection characteristic of photovoltaic power generation is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of solar power generation, in particular to a photovoltaic power station power sudden change early warning method based on ground meteorological station data. BACKGROUND

[0002] With the acceleration of global energy transformation, the installed capacity of photovoltaic power generation continues to grow, and has become an important source of electricity. However, photovoltaic power generation relies on solar irradiance, and its output power has significant intermittency and volatility, which brings great challenges to real-time power balance and safe and stable operation of the power system. Among them, the moving cloud layer blocking is the main factor leading to the rapid and rapid change of photovoltaic power, when the cloud layer drifts over the photovoltaic power station, the solar radiation received by the power station will decrease sharply in a few minutes or even tens of seconds (up to 80% of the rated power), forming a "sawtooth" power curve, which has a serious impact on the power grid. Therefore, the technology of early warning such power sudden change 5-15 minutes in advance has very high application value.

[0003] At present, the existing technologies for photovoltaic power sudden change early warning mainly have two types: one is the cloud movement prediction technology based on all-sky imager, and the other is the power prediction technology relying on numerical weather prediction. Among them, the most similar existing technology to the present application is "cloud amount real-time monitoring method based on all-sky imager", which is described in detail in patent document CN117808736A. The system hardware core is an all-sky imager, including a host equipped with a fisheye lens, a data receiving and processing unit and auxiliary components, which realizes cloud amount monitoring through image acquisition and exposure adjustment, ground object area removal, cloud amount detection and segmentation, thick and thin cloud separation, etc. to provide data support for photovoltaic power prediction.

[0004] However, the existing technology has obvious defects. The technology based on all-sky imager has high cost, needs to deploy special fisheye cameras and processing servers, and has high hardware investment and maintenance cost. In addition, in bad weather such as night, rain, snow and haze, the image quality is poor, the cloud layer cannot be effectively identified, and the system is prone to failure. At the same time, its image recognition algorithm is complex, is greatly affected by factors such as light angle and seasonal change, has poor stability, and time-consuming calculation may also weaken the early warning lead time. The technology relying on numerical weather prediction has coarse spatio-temporal resolution (usually kilometer level, hour level), cannot capture local and small-scale cloud cluster activities, and is difficult to realize short-term early warning. Therefore, there is an urgent need in the industry for a low-cost, high-reliability, high-time-efficiency photovoltaic power sudden change short-term early warning universal solution. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a photovoltaic power station power sudden change early warning method based on ground meteorological station data, which overcomes the defects of high cost, poor environmental adaptability and insufficient timeliness of the prior art.

[0006] To solve the above technical problems, the technical solution adopted by the present application is: The photovoltaic power station power sudden change early warning method based on ground meteorological station data comprises the following steps: Step1, data reading and preprocessing: read the latest total radiation data GHI and scattered radiation data DIF from the database cache of the photovoltaic power station central server in a timely manner, and perform validity test and smoothing filter processing to obtain clean and usable GHI t and DIF t data; Step2, calculate the key indicators: based on the data processed in Step1, calculate the scattering ratio K t and the total radiation drop rate R, wherein K t = DIF t / GHI t , when GHI t is less than 5W / m², K t is set to 0, R = (GHI t - GHI t-3 ) / 3, and GHI t-3 is the total radiation data 3 minutes ago; Step3, read dynamic threshold: read the currently effective total radiation drop rate threshold R th and the scattering ratio change rate threshold K th from the configuration file or special data table of the server; Step4, judgment logic: judge whether the total radiation drop rate R is less than R th , and calculate the change rate dK_dt = (K t - K t-2 ) / 2 of the scattering ratio in 2 minutes, K t-2 is the scattering ratio 2 minutes ago, judge whether dK_dt is greater than K th , if both conditions are met, trigger the early warning; Step5, early warning information generation and release: generate structured early warning information containing early warning unique identifier, trigger time, early warning type, expected power sudden change time and expected drop, and send it to the receiver through the preset message push interface.

[0007] In Step1 above, the validity test is specifically: check whether the read GHI and DIF data are within a reasonable physical range, the GHI data at night should be 0, and the GHI data during the day should not exceed the solar constant, if the data is out of range or null, it is considered as invalid data, discard this sampling, and wait for the next data reading.

[0008] 3. The photovoltaic power station power sudden change early warning method based on ground meteorological station data according to claim 1, characterized in that in Step 1, a moving average filter with a length of 3 is used for smoothing filtering, specifically: the average values of the three effective GHI data and the three effective DIF data at time t, t-1 and t-2 are calculated respectively, and the average values are used as the final GHI t and DIF t data at time t.

[0009] In Step 3, the dynamic threshold R th and K th are determined by historical data statistics method, and the specific steps include: Step 3.1, data preparation: collect the GHI, DIF data and historical power data of the photovoltaic power station every minute in the past 1-2 years, select the power drop events of more than 50% within 10 minutes from the power data, and mark the time points 3 minutes before the event as positive sample time points; Step 3.2, feature extraction: calculate the total radiation drop rate R and the scattering ratio change rate dK_dt of each positive sample time point, and randomly select a large number of time points as negative sample time points under clear and stable weather, and calculate the corresponding R and dK_dt; Step 3.3, distribution analysis: draw the distribution of positive and negative samples on a two-dimensional scatter plot, where X axis is R and Y axis is dK_dt, and the positive sample set is distributed in the lower left quadrant; Step 3.4, threshold determination: determine the R th and K th that can distinguish the positive and negative sample sets by machine learning classification boundary determination method or selecting the highest accuracy combination, and write them into the configuration file or special data table of the server.

[0010] In Step 5, the structured early warning information adopts JSON format, which specifically includes early warning unique identifier alert_id, early warning trigger time timestamp, early warning type alert_type, alert_type takes the value of "power sudden drop early warning", expected power sudden change time expected_time, expected_time is the current time plus 5 minutes, and expected power drop amplitude expected_impact, which is determined according to the corresponding relationship between R and dK_dt and power drop amplitude.

[0011] In Step 5, the preset message push interface includes HTTP API interface, MQTT client interface and SMS gateway interface, which can select the corresponding interface to send early warning information according to the needs of the receiver.

[0012] The Step 5 above further comprises a step of updating a historical data queue: storing the GHI t , DIF t , K t data and the corresponding time stamp of the current processing into the historical data queue for subsequent calculation of the total radiation decline rate R and the scattering ratio change rate dK_dt.

[0013] The historical data queue above adopts a first-in-first-out storage rule, and only retains historical data of the last 30 days, and when the data storage time exceeds 30 days, the earliest historical data is automatically deleted, so as to ensure that the data storage capacity is reasonable.

[0014] The time interval of data reading in the Step 1 above can be configured, and the range is 1-5 minutes, and the default time interval is 1 minute, which can be adjusted according to the real-time demand of the photovoltaic power station on early warning.

[0015] When it is judged that the early warning is not triggered in the Step 4 above, the Step 1 is returned, and the next data reading and preprocessing are continued, so as to realize continuous monitoring of the power sudden change of the photovoltaic power station.

[0016] The photovoltaic power station power sudden change early warning method based on ground meteorological station data mentioned in the application takes the scattering ratio mutation as the core criterion, constructs a double-condition judgment logic in combination with the total radiation decline rate, utilizes the existing meteorological data of the photovoltaic power station, does not need to add new hardware, and has low cost; is based on radiation measurement, is not affected by severe weather, and has strong environmental adaptability; the algorithm only involves simple arithmetic operation and logical judgment, and has high response real-time performance; the early warning mechanism is based on clear atmospheric physics principle, has high accuracy, can effectively provide a buffer time for power grid dispatching and energy storage adjustment, and improves the friendly grid connection characteristics of photovoltaic power generation.

[0017] The application has the following beneficial effects: 1. Low cost, easy to implement The prior art scheme depends on expensive special fisheye cameras and complex image processing servers, and the hardware investment and software development cost are high. The application does not need to add new hardware devices, and completely utilizes the existing and necessary standard meteorological station (already equipped with GHI and DIF sensors) and SCADA system of the photovoltaic power station. Therefore, the application realizes "zero marginal hardware cost" of the early warning function, and has the advantage of popularization and application.

[0018] 2. Strong environmental adaptability and high reliability The prior art is based on optical imaging, and in low light or bad weather such as night, rain, snow, haze, the image quality decreases, leading to system failure and insufficient reliability. The present application is based on radiation measurement, and its effectiveness is not limited by light conditions and weather phenomena. Whether it is day or night (it can automatically sleep by judging the GHI value), whether it is sunny or bad weather, the data acquisition and processing logic can run stably, providing 7x24 hours of uninterrupted stable monitoring capability, and the environmental robustness will be better than the prior art.

[0019] 3. The algorithm is simple and efficient, and the response is real-time The prior art requires complex image preprocessing, cloud layer segmentation, motion vector estimation and a large amount of calculation, which takes a long time and may affect the timeliness of the early warning. The core algorithm of the present application only involves simple arithmetic operations (division and difference) and logical judgments, and the calculation overhead is extremely small. This makes the system able to complete all analysis within seconds, faster response, and can gain more valuable early warning time for the power grid.

[0020] 4. The early warning mechanism is direct, the physical meaning is clear, and the accuracy is high The prior art tries to indirectly infer the change of radiation by "looking" at the cloud and predicting its path, which is long and has large cumulative error. The present application directly monitors the mutation characteristics of the key parameters (GHI and DIF) of solar radiation itself. The criterion of "scattering ratio sudden increase" is derived from the clear atmospheric physics principle that "translucent cloud layer causes scattered radiation to increase before total radiation decreases", and the mechanism is direct. Combined with the judgment of the total radiation decline rate, false alarms (such as solar eclipse, instantaneous obstruction) are effectively avoided, making the accuracy of the early warning significantly higher than the image recognition-based method.

[0021] 5. Easy maintenance, low operating cost The lens of the all-sky imager needs to be cleaned and maintained regularly, and the algorithm model needs to be continuously optimized to adapt to different light conditions, which has high maintenance cost. The radiation sensor relied on by the present application is already a routine work of the power station, and the algorithm runs stably without frequent adjustment, almost without additional maintenance burden and operating cost. BRIEF DESCRIPTION OF DRAWINGS

[0022] The present application will be further described below in conjunction with the drawings and examples: Figure 1 The overall flowchart of the photovoltaic power station power sudden change early warning method of the present application. DETAILED DESCRIPTION

[0023] The technical solutions of the present application will be described in detail below in conjunction with the drawings and examples.

[0024] The photovoltaic power station power sudden change early warning method based on ground meteorological station data includes the following steps: Step1, data reading and preprocessing: read the latest total radiation data GHI and scattered radiation data DIF from the database cache of the photovoltaic power station central server in a timely manner, and perform validity test and smoothing filter processing to obtain clean and usable GHI t and DIF t data; Step2, calculate key indicators: based on the data processed in Step1, calculate the scattering ratio K t and the total radiation decline rate R, wherein K t = DIF t / GHI t , when GHI t is less than 5W / m², K t is set to 0, R = (GHI t - GHI t-3 ) / 3, GHI t-3 is the total radiation data 3 minutes ago; Step3, read dynamic threshold: read the currently effective total radiation decline rate threshold R th and the scattering ratio change rate threshold K th from the server's configuration file or special data table; Step4, judgment logic: judge whether the total radiation decline rate R is less than R th , and calculate the change rate of the scattering ratio in 2 minutes dK_dt = (K t - K t-2 ) / 2, K t-2 is the scattering ratio 2 minutes ago, judge whether dK_dt is greater than K th , if both conditions are met, trigger the warning; Step5, warning information generation and release: generate structured warning information containing warning unique identifier, trigger time, warning type, expected power sudden change time and expected drop, and send it to the receiver through the preset message push interface.

[0025] In Step1 above, the validity test is specifically: check whether the read GHI and DIF data are within a reasonable physical range, the GHI data at night should be 0, and the GHI data during the day should not exceed the solar constant, if the data is out of range or null, it is considered as invalid data, discard this sampling, wait for the next data reading.

[0026] 3. The photovoltaic power station power sudden change early warning method based on ground meteorological station data according to claim 1, characterized in that in Step 1, a moving average filter with a length of 3 is used for smoothing filtering, specifically: the average values of the three effective GHI data and the three effective DIF data at time t, t-1 and t-2 are calculated respectively, and the average values are used as the final GHI t and DIF t data at time t.

[0027] In Step 3, the dynamic threshold R th and K th are determined by historical data statistics method, and the specific steps include: Step 3.1, data preparation: collect the GHI, DIF data and historical power data of the photovoltaic power station every minute in the past 1-2 years, select the power drop events of more than 50% within 10 minutes from the power data, and mark the time points 3 minutes before the event as positive sample time points; Step 3.2, feature extraction: calculate the total radiation drop rate R and the scattering ratio change rate dK_dt of each positive sample time point, and randomly select a large number of time points as negative sample time points under clear and stable weather, and calculate the corresponding R and dK_dt; Step 3.3, distribution analysis: draw the distribution of positive and negative samples on a two-dimensional scatter plot, where X axis is R and Y axis is dK_dt, and the positive sample set is distributed in the lower left quadrant; Step 34, threshold determination: determine the R th and K th that can distinguish the positive and negative sample sets by machine learning classification boundary determination method or selecting the highest accuracy combination, and write them into the configuration file or special data table of the server.

[0028] In Step 5, the structured early warning information adopts JSON format, which specifically includes early warning unique identifier alert_id, early warning trigger time timestamp, early warning type alert_type, alert_type takes the value of "power sudden drop early warning", expected power sudden change time expected_time, expected_time is the current time plus 5 minutes, and expected power drop amplitude expected_impact, which is determined according to the corresponding relationship between R and dK_dt and power drop amplitude.

[0029] In Step 5, the preset message push interface includes HTTP API interface, MQTT client interface and SMS gateway interface, which can select the corresponding interface to send early warning information according to the needs of the receiver.

[0030] Step 5 above also includes the step of updating the historical data queue: updating the GHI data processed in this operation. t DIF t K t The data and corresponding timestamps are stored in the historical data queue, which is used to call historical data when calculating the total radiation decrease rate R and the scattering ratio change rate dK_dt.

[0031] The aforementioned historical data queue adopts a first-in, first-out (FIFO) storage rule, retaining only the most recent 30 days of historical data. When the data storage time exceeds 30 days, the oldest historical data is automatically deleted to ensure reasonable data storage capacity.

[0032] The data reading time interval in Step 1 above is configurable, ranging from 1 to 5 minutes, with a default time interval of 1 minute. This interval can be adjusted according to the real-time early warning requirements of the photovoltaic power station.

[0033] If no warning is triggered in Step 4 above, return to Step 1 to continue the next data reading and preprocessing, so as to achieve continuous monitoring of sudden power changes in photovoltaic power plants.

[0034] Example 1: like Figure 1 As shown in the flowchart, each step of the technical solution of the present invention will be described in detail: Step S1: Data Reading and Preprocessing Function: Retrieves raw, potentially noisy data from a real-time database and processes it into clean, usable data.

[0035] Specific technical implementation: 1. Reading: The algorithm program periodically (e.g., every minute) reads the latest GHI and DIF data from the central server's database cache and adds a timestamp t.

[0036] 2. Validity check: Check whether the data is within a reasonable physical range (e.g., GHI is 0 at night, and should not exceed the solar constant during the day). If the data is outside the range or is null, it is considered invalid data, the current sample is discarded, and the next sample is awaited.

[0037] 3. Smoothing Filter: To suppress transient noise interference, a moving average filter of length 3 is used. That is, the average of three valid data values ​​at times t, t-1, and t-2 is taken as the final adopted values ​​GHI_t and DIF_t at time t. This is a simple digital signal processing technique commonly used in this field.

[0038] Step S2: Calculate key performance indicators K and R Function: Extract features from raw data for decision-making.

[0039] Specific technical implementation: 1. Calculate the scattering ratio K t : Perform a division operation K t = DIF t / GHI t . To prevent division by zero errors, when GHI t is less than a very small value (such as 5 W / m²), directly set K t to 0.

[0040] 2. Calculate the total radiation decline rate R: 1) Retrieve the total radiation value GHI_t-3 three minutes ago (time t - 3) from the historical data queue.

[0041] 2) Perform the calculation R = (GHI t - GHI t-3 ) / 3. The unit is W / m² / minute. This calculation is a simple difference operation.

[0042] Step S3: Read the dynamic threshold R th and K th Function: Obtain the reference value required for judgment, and this reference value can be adjusted adaptively.

[0043] Specific technical implementation: 1) The threshold is not fixed in the code, but is stored as a configurable parameter in the server's configuration file or a dedicated data table.

[0044] 2) The algorithm reads the currently effective threshold R th (such as -50) and K th (such as +0.15). These thresholds can be updated through the background management program according to the method in Step S3.

[0045] Step S4: Judgment logic Function: Make a decision on whether to give an early warning based on key indicators and thresholds.

[0046] Specific technical implementation: This is a logic judgment unit, and its execution process is represented by the following pseudocode: if (R<R_th): # Condition 1: Is the total radiation decreasing rapidly? dK = K_t - K_t-2 # Calculate the change in the scattering ratio within 2 minutes dK_dt = dK / 2 # Calculate the change rate if (dK_dt>K_th): # Condition 2: Does the scattering ratio increase sharply? return True # If conditions 1 and 2 are both met, trigger an alert. return False # Otherwise, do not trigger. This judgment logic is implemented by the server CPU executing condition judgment instructions.

[0047] Step S5: Triggering Early Warning and Information Generation Function: When the conditions are met, generate and send out structured early warning information.

[0048] Specific technical implementation: 1. Information Generation: The warning information is a data packet in structure or JSON format, containing: alert_id: A unique identifier for alerts.

[0049] timestamp: The time when the alert was triggered.

[0050] alert_type: "Power drop warning".

[0051] expected_time: Calculates the current time plus T (e.g., T = 5 minutes).

[0052] expected_impact: Based on historical statistics, match the power drop range (e.g., "70%-85%) corresponding to the current R and dK_dt values ​​from the lookup table.

[0053] 2. Publish information: Call the existing message push service interface on the server (such as HTTP API, MQTT client, SMS gateway interface) to send the warning information data packet to the intended recipient.

[0054] Example 2: To better understand and implement this invention, a specific numerical embodiment is provided below.

[0055] Time: 12:00 noon on a certain day.

[0056] data: t-3 time (11:57): GHI t-3 = 800 W / m², DIF t-3 = 100 W / m²→K t-3 = 100 / 800 = 0.125 Time t (12:00): GHI t = 600 W / m², DIF t= 250 W / m² → K t = 250 / 600 ≈ 0.417 calculate: The rate of descent R = (600 - 800) / 3 ≈ -66.7 W / m² / min The rate of change of scattering ratio ΔK / Δt = (0.417 - 0.125) / 2 = 0.146 / minute judge: Preset threshold R_th = -50, K th = +0.12.

[0057] Judgment 1: R (-66.7) <R th (-50) → Established.

[0058] Judgment 2: ΔK / Δt (0.146) > K th (0.12) → Found.

[0059] Result: The system immediately triggered an early warning at 12:00, generating the message: "[Power Drop Warning] It is expected that the power of this station will drop sharply around 12:05, with an estimated drop of 70%-85%. Please pay attention, relevant units." Specific technical solutions for threshold determination method Function: Enables threshold values ​​to adapt to different power plants and seasons, maintaining the accuracy of early warnings.

[0060] Specific technical solutions (taking historical data statistical method as an example): 1. Data Preparation: Collect GHI, DIF, and historical power data of the power plant every minute over the past 1-2 years. From the power data, find "sudden drop events" where the power drops by more than 50% in a short period of time (e.g., 10 minutes), and mark the time point 3 minutes before its occurrence as the "positive sample time point".

[0061] 2. Feature Extraction: For each "positive sample time point", calculate its R and ΔK / Δt values. Simultaneously, under clear and stable weather conditions, randomly select a large number of "negative sample time points" and calculate the same indicators.

[0062] 3. Distribution Analysis: Plot the distribution of positive and negative samples on a two-dimensional scatter plot (X-axis is R, Y-axis is ΔK / Δt). It will be observed that positive samples are concentrated in the lower left quadrant (R is negative, ΔK / Δt is positive).

[0063] 4. Threshold Determination: Select the R_th and K_th values ​​that best distinguish between positive and negative sample sets. For example, this can be done using a "classification boundary" method in machine learning, or simply by selecting the combination that maximizes the "accuracy". Finally, write the determined thresholds into the system configuration.

Claims

1. A method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data, characterized in that, Includes the following steps: Step 1: Data Reading and Preprocessing: The latest total radiation data (GHI) and diffuse radiation data (DIF) are periodically read from the database cache of the photovoltaic power station's central server. Validity checks and smoothing filters are then performed to obtain a clean and usable GHI. t and DIF t data; Step 2: Calculate key indicators: Based on the data processed in Step 1, calculate the scattering ratio K. t And the total radiation decrease rate R, where K t = DIF t / GHI t When GHI t When it is less than 5W / m², K t Set to 0, R = (GHI) t - GHI t-3 ) / 3, GHI t-3 This is the total radiation data from 3 minutes ago; Step 3: Read the dynamic threshold: Read the currently effective total radiation decrease rate threshold R from the server's configuration file or dedicated data table. th and the threshold of the rate of change of scattering ratio K th ; Step 4: Judgment Logic: Determine whether the total radiation decrease rate R is less than R0. th Simultaneously calculate the rate of change of the scattering ratio dK_dt = (K t - K t-2 ) / 2, K t-2 Given the scattering ratio from 2 minutes ago, determine if dK_dt is greater than K. th If both conditions are met, an alert will be triggered. Step 5: Generation and release of early warning information: Generate structured early warning information containing a unique early warning identifier, trigger time, early warning type, expected power change time and expected decrease, and send it to the recipient through a preset message push interface.

2. The method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data according to claim 1, characterized in that, In Step 1, the validity check specifically involves checking whether the read GHI and DIF data are within a reasonable physical range. The nighttime GHI data should be 0, and the daytime GHI data should not exceed the solar constant. If the data is outside the range or is empty, it is considered invalid data, and the current sample is discarded, waiting for the next data read.

3. The method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data according to claim 1, characterized in that, In Step 1, the smoothing filter uses a moving average filter of length 3. Specifically, it calculates the average of three valid GHI data points and three valid DIF data points at times t, t-1, and t-2, and uses this average as the final GHI value used at time t. t and DIF t data.

4. The method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data according to claim 1, characterized in that, In Step 3, the dynamic threshold R th and K th This was determined using historical data statistical methods, and the specific steps included: Step 3.1 Data preparation: Collect GHI, DIF data and historical power data of photovoltaic power plants every minute over the past 1-2 years, screen out sudden drop events with a power drop of more than 50% within 10 minutes from the power data, and mark the time point 3 minutes before the event as the positive sample time point; Step 3.2 Feature extraction: Calculate the total radiation decrease rate R and the scattering ratio change rate dK_dt for each positive sample time point. At the same time, under clear and stable weather conditions, randomly select a set number of time points as negative sample time points and calculate their corresponding R and dK_dt. Step 3.3 Distribution Analysis: Plot the distribution of positive and negative samples on a two-dimensional scatter plot, where the X-axis is R and the Y-axis is dK_dt, with positive samples concentrated in the lower left quadrant; Step 34: Threshold Determination: Determine the R-value that can distinguish between positive and negative sample sets by using machine learning classification boundary determination methods or selecting the combination with the highest accuracy. th and K th And write it into the server's configuration file or a dedicated data table.

5. The method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data according to claim 1, characterized in that, In Step 5, the structured warning information adopts JSON format, specifically including the warning unique identifier alert_id, the warning trigger time timestamp, the warning type alert_type, the value of alert_type is "power drop warning", the expected power drop time expected_time, expected_time is the current time plus 5 minutes, and the expected power drop expected_impact, expected_impact is determined according to the correspondence between historical statistics of R and dK_dt and the power drop.

6. The method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data according to claim 1, characterized in that, In Step 5, the preset message push interfaces include HTTP API interface, MQTT client interface and SMS gateway interface. The recipient can select the corresponding interface to send the warning information according to their needs.

7. The method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data according to claim 1, characterized in that, Step 5 is followed by a step of updating the historical data queue: updating the GHI data processed in this step... t DIF t K t The data and corresponding timestamps are stored in the historical data queue, which is used to call historical data when calculating the total radiation decrease rate R and the scattering ratio change rate dK_dt.

8. The method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data according to claim 7, characterized in that, The historical data queue adopts a first-in-first-out storage rule, retaining only the most recent 30 days of historical data. When the data storage time exceeds 30 days, the oldest historical data is automatically deleted to ensure reasonable data storage capacity.

9. The method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data according to claim 1, characterized in that, The data reading time interval in Step 1 is configurable, ranging from 1 to 5 minutes, with a default time interval of 1 minute, which can be adjusted according to the real-time early warning requirements of the photovoltaic power station.

10. The method for early warning of sudden power changes in photovoltaic power plants based on ground meteorological station data according to claim 1, characterized in that, In Step 4, if it is determined that no warning has been triggered, return to Step 1 and continue to read and preprocess the data for the next time to achieve continuous monitoring of sudden changes in the power of the photovoltaic power station.

Images