Photovoltaic power generation data encryption protection method based on local differential privacy

Abstract

The invention discloses a photovoltaic power generation data encryption protection method based on local differential privacy, and the method comprises the steps: firstly building a privacy parameter configuration mechanism, generating a random signal through employing a Laplacian mechanism, enabling the signal to serve as an invisible watermark at a data source, and enabling the invisible watermark to be embedded into original data, and obtaining protection data; then the system automatically records key parameters in the watermark embedding process to form a complete embedded evidence chain; the detection end compares suspected data with original reference data, calculates an actual signal sequence, performs integrity verification and abnormal value detection on the extracted signal sequence, and performs watermark detection by using a triple distribution similarity detection mechanism to obtain a comprehensive similarity score and a watermark confidence level; and finally generating a detection technology report. According to the method, privacy protection is realized at a data source, the security is high, the core characteristics of photovoltaic data are kept, data traceability and copyright protection are realized, and technical support is provided for constructing a safe, efficient and intelligent novel power system.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic data encryption and protection technology, and in particular to a method for encrypting and protecting photovoltaic power generation data based on local differential privacy. Background Technology

[0002] With the deepening of the "dual-carbon" goals and the accelerated construction of new power systems, the photovoltaic industry has experienced explosive growth. Photovoltaic power generation data, including power output, operating status, and environmental parameters, is not only a crucial basis for grid dispatch but also a core asset for assessing power generation efficiency, predicting power output, and diagnosing faults. This data is characterized by high precision, strong time-series characteristics, and massive volume; its completeness and accuracy directly affect grid security, power trading settlement, and equipment status assessment. However, photovoltaic power generation data faces severe security challenges during its collection, transmission, storage, and sharing.

[0003] (1) Data leakage risk: Photovoltaic data contains sensitive information such as users' electricity consumption behavior and geographical location. Once leaked, it may infringe on users' privacy.

[0004] (2) Data tampering threat: Malicious attackers may tamper with power generation data, affecting grid dispatch and electricity billing;

[0005] (3) Lack of copyright protection: In data trading and sharing scenarios, there is a lack of effective technical means to track data sources and prevent unauthorized dissemination;

[0006] (4) The contradiction between privacy and utility: Although traditional encryption technology can protect privacy, it seriously hinders the analysis and use of data.

[0007] Currently, the main technical solutions for photovoltaic data protection include traditional encryption technology, traditional digital watermarking technology, traditional differential privacy technology, and localized differential privacy technology. Traditional encryption technologies use algorithms such as AES and RSA to encrypt data as a whole. While offering high security, the encrypted data becomes "ciphertext," completely losing its usability and making it unsuitable for direct use in data analysis, machine learning, and other applications. This severely conflicts with the business requirements of real-time monitoring and analysis of photovoltaic data. Traditional digital watermarking technology embeds visible or invisible identification information into the data, primarily for copyright protection, but suffers from insufficient privacy protection, poor robustness, and weak security. Traditional differential privacy technology provides strict mathematical privacy guarantees by adding random signals to query results, but it is mainly suitable for centralized data processing scenarios. In photovoltaic environments, data is scattered across tens of thousands of photovoltaic sites, and the scheme of collecting data centrally and then adding signals has problems such as single point of failure risk, high communication overhead, and poor timeliness. Localized differential privacy technology perturbs user data locally before sending it to the aggregator, rather than directly transmitting the original data, thereby avoiding the leakage of sensitive information. It is mainly used for simple statistical queries, but has significant limitations in photovoltaic data protection, such as crude signal control, lack of watermarking mechanisms, and limited detection methods. How to achieve a comprehensive solution in photovoltaic scenarios that provides strict privacy protection, maintains high data availability, and has data traceability and copyright protection capabilities is an urgent problem to be solved in the encryption and protection of photovoltaic power generation data. Summary of the Invention

[0008] The purpose of this invention is to provide a photovoltaic power generation data encryption protection method that can balance privacy protection and data availability, support data traceability and copyright protection, and has strong resistance to privacy attacks and high security.

[0009] The technical solution to achieve the purpose of this invention is: a method for encrypting and protecting photovoltaic power generation data based on local differential privacy, comprising the following steps:

[0010] Step 1: Establish a privacy parameter configuration mechanism;

[0011] Step 2: Generate a random signal using the Laplace mechanism;

[0012] Step 3: Embed the signal as an invisible watermark into the original data at the data source to obtain protected data;

[0013] Step 4: The system automatically records key parameters during the watermark embedding process, forming a complete embedding evidence chain;

[0014] Step 5: The detection end calculates the actual signal sequence by comparing the suspected data with the original reference data, and performs integrity verification and outlier detection on the extracted signal sequence;

[0015] Step 6: Use the triple distribution similarity test mechanism to detect the watermark and obtain the comprehensive similarity score and watermark confidence level;

[0016] Step 7: Generate a testing technical report and complete the encryption protection of photovoltaic power generation data.

[0017] Furthermore, the mechanism for establishing privacy parameter configuration described in step 1 is as follows:

[0018] Based on the sensitivity characteristics of photovoltaic data and the accuracy requirements of business operations, the optimal privacy parameters are dynamically calculated. The system presets a privacy budget of ε=0.8 and a data sensitivity of Δf=1.0. Through the precise calculation of the Laplace signal scale parameter λ=Δf / ε, it ensures that data utility is maintained while providing strict privacy protection.

[0019] Furthermore, the generation of random signals using the Laplace mechanism described in step 2 is as follows:

[0020] Step 2.1: Generate a random signal using the Laplace mechanism. The signal follows a Laplace distribution Laplace(0,λ). Calculate the theoretical quantiles and dynamically adjust the signal range to ensure that the absolute value of the signal does not exceed the set value at a 99% confidence level.

[0021] Step 2.2: The system monitors the statistical characteristics of the generated signal in real time. When the theoretical signal limit exceeds the preset threshold, the signal truncation function is automatically activated to cut the signal value that exceeds the range to the allowable range. At the same time, the truncation ratio is recorded for quality assessment.

[0022] Furthermore, step 3 involves embedding the signal as an invisible watermark into the original data at the data source to obtain protected data, as detailed below:

[0023] For the power generation data of each photovoltaic site, the system generates a corresponding Laplace signal, which is then embedded as an invisible watermark into the original data to obtain protected data.

[0024] Protected data = raw data + controlled signals

[0025] The raw data refers to the actual power generation data collected by the photovoltaic site; the controlled signal refers to the random signal generated by the Laplace mechanism and controlled by range; the protected data refers to the result after the signal is embedded, which is both privacy-preserving data that satisfies local differential privacy and copyright identification data with embedded invisible watermark.

[0026] The embedding process ensures that the temporal and statistical characteristics of the data are preserved, and the correlation coefficient between the watermarked data and the original data is maintained above 0.98.

[0027] Furthermore, the key parameters in the watermark embedding process described in step 4 include privacy budget, signal scale, actual signal statistics, and embedding timestamp.

[0028] Furthermore, the calculation formula for the actual signal sequence mentioned in step 5 is as follows:

[0029] Actual signal = Suspected data - Original data

[0030] The above formula describes how to extract possible watermarks, i.e., signal sequences, from the data to be verified in the detection segment;

[0031] The suspected data refers to data that is discovered during circulation or use and whose source and copyright need to be verified.

[0032] The raw data refers to the real data used as a comparison benchmark, which has not been processed in any way;

[0033] The actual signal is the signal sequence extracted through differential calculation; if the suspected data is the protected data obtained in step 3, then the extracted actual signal and the controlled signal during embedding are consistent in statistical distribution.

[0034] Furthermore, step 6 involves using a triple distribution similarity test mechanism to detect the watermark, obtaining a comprehensive similarity score and a watermark confidence level, as detailed below:

[0035] Step 6.1: Use the triple distribution similarity test mechanism to detect watermarks.

[0036] First step: Consistency test of KS distribution

[0037] The Kolmogorov-Smirnov test is used to compare the consistency between the actual signal distribution and the theoretical Laplace distribution. The KS statistic and the corresponding p-value are calculated, and the statistic is converted into a similarity score.

[0038] Similarity = 1 - KS statistic

[0039] The significance level was set at 1%, and a p-value greater than 0.01 was considered a consistent distribution.

[0040] Second step: Variance consistency test

[0041] By comparing the consistency between the actual signal variance and the theoretical variance, and calculating the variance ratio, a similarity assessment can be obtained.

[0042] Similarity = 1 - |variance ratio - 1|

[0043] The above formula is used to evaluate whether the dispersion of the actual signal is consistent with the expected theory;

[0044] Variance ratio = Actual signal variance / Theoretical variance, where the theoretical variance of the Laplace mechanism is 2λ²; when the acceptable range of variance ratio is set between 0.5 and 1.5, it is considered that the actual signal variance and the theoretical variance are consistent; the acceptable range of variance ratio is a tolerance interval set based on business experience.

[0045] The variance ratio - 1 calculates the absolute mutual error between the actual variance and the theoretical variance; the closer the variance ratio is to 1, the smaller this error is.

[0046] The similarity score is calculated as 1 - |variance ratio - 1|. The closer the score is to 1, the better the variance consistency.

[0047] Third step: Quantile matching test

[0048] Key quantiles of 5%, 25%, 50%, 75%, and 95% were selected. The matching degree between the actual signal quantiles and the theoretical quantiles was compared, the average quantile error was calculated, and the similarity score was obtained.

[0049] Similarity = 1 - Average Error

[0050] The above formula is used to evaluate the consistency between the actual signal distribution shape and the theoretical signal distribution;

[0051] The average error refers to the average of the absolute errors calculated at selected key points at 5%, 25%, 50%, 75%, and 95%, respectively, by subtracting the theoretical signal quantile from the actual signal quantile.

[0052] Subtract the average error to obtain the similarity score. The higher the score, the more similar the distribution shape. Set an error threshold to ensure the consistency of the distribution shape.

[0053] Step 6.2: Based on the results of the triple test, a weighted fusion algorithm is used to calculate the comprehensive similarity score, assigning 40% weight to the KS test, 35% weight to the variance test, and 25% weight to the quantile test to reflect the differences in importance of each test.

[0054] Step 6.3: Set the detection threshold to 0.7. When the overall similarity reaches or exceeds the threshold, the watermark is determined to exist. The confidence level is divided according to the score range: above 0.8 is high confidence, 0.6 to 0.8 is medium confidence, and below 0.6 is low confidence.

[0055] Furthermore, in step 6.1, the watermark detection using the triple distribution similarity test mechanism is designed with a partial data comparison detection mechanism: when there are partial missing original parameter data or suspected data, signal extraction and subsequent triple distribution test are performed only based on the existing data points.

[0056] The system does not require the dataset for comparison to be 100% complete, but rather performs calculations based on existing data; when the data missing rate is within 10%, the detection accuracy based on partial data can still remain above 90%.

[0057] Furthermore, the detection threshold mentioned in step 6.3 is dynamically adjusted based on data quality and the detection environment:

[0058] Standard thresholds are used when the data is complete, the sequence is stable, the computing power is high and the latency is low, and the data quality is high and the environment is stable.

[0059] In cases where the data missing rate is greater than 10%, the sequence fluctuates severely, computing power is limited, and the data quality is low due to high latency and complex environment, the threshold requirements can be relaxed to balance detection sensitivity and false alarm rate.

[0060] Furthermore, the generation of the detection technology report in step 7 is as follows:

[0061] Each test generates a detailed technical report, including signal statistical analysis, detailed results of each test item, comprehensive similarity score, watermark existence judgment and confidence level assessment; the report adopts a standardized format and has legal evidentiary effect.

[0062] Compared with the prior art, the present invention has the following significant advantages: (1) It achieves privacy protection at the source of data, namely each photovoltaic site, and avoids leakage of original data; (2) The processed data still retains the core characteristics of photovoltaic data, supporting accurate power generation prediction and grid dispatch; (3) It embeds imperceptible but detectable identification information, realizing data traceability and copyright protection; (4) It provides strict mathematical privacy guarantee, which can resist various privacy attacks and has high security; (5) It adapts to the high precision and time sequence characteristics of photovoltaic power generation data, providing key technical support for building a safe, efficient and intelligent new power system. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating the photovoltaic power generation data encryption protection method based on local differential privacy according to the present invention.

[0064] Figure 2 This is a comparison chart of data distribution before and after watermarking in an embodiment of the present invention.

[0065] Figure 3 This is a comparison chart of data quality before and after watermarking in an embodiment of the present invention. Detailed Implementation

[0066] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0067] like Figure 1As shown, the present invention provides a method for encrypting and protecting photovoltaic power generation data based on local differential privacy, comprising the following steps:

[0068] Step 1: Establish a privacy parameter configuration mechanism;

[0069] Step 2: Generate a random signal using the Laplace mechanism;

[0070] Step 3: Embed the signal as an invisible watermark into the original data at the data source to obtain protected data;

[0071] Step 4: The system automatically records key parameters during the watermark embedding process, forming a complete embedding evidence chain;

[0072] Step 5: The detection end calculates the actual signal sequence by comparing the suspected data with the original reference data, and performs integrity verification and outlier detection on the extracted signal sequence;

[0073] Step 6: Use the triple distribution similarity test mechanism to detect the watermark and obtain the comprehensive similarity score and watermark confidence level;

[0074] Step 7: Generate a testing technical report and complete the encryption protection of photovoltaic power generation data.

[0075] The detection report is crucial and serves as proof for this method to achieve copyright protection and data traceability. It is not only a summary of the watermark detection results but also a manifestation of the entire closed-loop encryption system. In the event of a data dispute, the report is used to explain the data source, verify data integrity, and assess the strength of protection, thus directly supporting the copyright protection and liability determination objectives within the "encryption protection" framework.

[0076] As a specific example, the establishment of the privacy parameter configuration mechanism described in step 1 is as follows:

[0077] Based on the sensitivity characteristics of photovoltaic data and the accuracy requirements of business operations, the optimal privacy parameters are dynamically calculated. The system presets a privacy budget of ε=0.8 and a data sensitivity of Δf=1.0. Through the precise calculation of the Laplace signal scale parameter λ=Δf / ε, it ensures that data utility is maintained while providing strict privacy protection.

[0078] A privacy budget of ε=0.8 means that even with maximum background knowledge, attackers cannot infer the true power generation information of any individual photovoltaic site by observing the processed data. The privacy protection strength reaches industrial-grade security standards, effectively resisting various privacy attack methods, including background knowledge attacks and re-identification attacks. While providing strict privacy protection, data accuracy loss is controlled within an acceptable range. The processed photovoltaic data still maintains two decimal places of accuracy, and the statistical feature retention rate of power generation data exceeds 95%, fully meeting the accuracy requirements for power trading settlement and power generation statistics. The protected data performs excellently in critical business scenarios. Compared with traditional unprotected data, the prediction error of power generation prediction only increases by 0.5%; the correlation of grid dispatch support data remains above 0.98, without affecting real-time dispatch decisions; and the anomaly detection sensitivity of fault detection capability remains above 92%.

[0079] As a specific example, step 2 involves generating a random signal using the Laplace mechanism, as detailed below:

[0080] Step 2.1: Generate a random signal using the Laplace mechanism. The signal follows a Laplace distribution Laplace(0,λ). Calculate the theoretical quantiles and dynamically adjust the signal range to ensure that the absolute value of the signal does not exceed the set value at a 99% confidence level.

[0081] Step 2.2: The system monitors the statistical characteristics of the generated signal in real time. When the theoretical signal limit exceeds the preset threshold, the signal truncation function is automatically activated to cut the signal value that exceeds the range to the allowable range. At the same time, the truncation ratio is recorded for quality assessment.

[0082] As a specific example, step 3 involves embedding the signal as an invisible watermark into the original data at the data source to obtain protected data, as detailed below:

[0083] For the power generation data of each photovoltaic site, the system generates a corresponding Laplace signal, which is then embedded as an invisible watermark into the original data to obtain protected data.

[0084] Protected data = raw data + controlled signals

[0085] The raw data refers to the actual power generation data collected by the photovoltaic site; the controlled signal refers to the random signal generated by the Laplace mechanism and controlled by range; the protected data refers to the result after the signal is embedded, which is both privacy-preserving data that satisfies local differential privacy and copyright identification data with embedded invisible watermark.

[0086] The embedding process ensures that the temporal and statistical characteristics of the data are preserved, and the correlation coefficient between the watermarked data and the original data is maintained above 0.98.

[0087] Privacy protection is implemented directly at the data source. Each photovoltaic site completes privacy processing before the data leaves the local area, completely eliminating the risk of privacy leakage during data transmission. Compared with traditional centralized privacy protection, it avoids the single point of failure problem of data aggregation nodes, and the reliability of privacy protection is improved to over 99.9%.

[0088] As a specific example, the key parameters in the watermark embedding process described in step 4 include privacy budget, signal scale, actual signal statistics, and embedding timestamp, forming a complete embedding evidence chain;

[0089] As a specific example, the formula for calculating the actual signal sequence in step 5 is as follows:

[0090] Actual signal = Suspected data - Original data

[0091] The above formula describes how to extract possible watermarks, i.e., signal sequences, from the data to be verified in the detection segment;

[0092] The suspected data refers to data that is discovered during circulation or use and whose source and copyright need to be verified.

[0093] The raw data refers to the real data used as a comparison benchmark, which has not been processed in any way;

[0094] The actual signal is the signal sequence extracted through differential calculation; if the suspected data is the protected data obtained in step 3, then the extracted actual signal and the controlled signal during embedding are consistent in statistical distribution.

[0095] As a specific example, step 6 describes using a triple distribution similarity test mechanism to detect watermarks, obtaining a comprehensive similarity score and watermark confidence level, as follows:

[0096] Step 6.1: Use the triple distribution similarity test mechanism to detect watermarks.

[0097] First step: Consistency test of KS distribution

[0098] The Kolmogorov-Smirnov test is used to compare the consistency between the actual signal distribution and the theoretical Laplace distribution. The KS statistic and the corresponding p-value are calculated, and the statistic is converted into a similarity score.

[0099] Similarity = 1 - KS statistic

[0100] The significance level was set at 1%, and a p-value greater than 0.01 was considered a consistent distribution.

[0101] Second step: Variance consistency test

[0102] By comparing the consistency between the actual signal variance and the theoretical variance, and calculating the variance ratio, a similarity assessment can be obtained.

[0103] Similarity = 1 - |variance ratio - 1|

[0104] The above formula is used to evaluate whether the dispersion of the actual signal is consistent with the expected theory;

[0105] Variance ratio = Actual signal variance / Theoretical variance, where the theoretical variance of the Laplace mechanism is 2λ²; when the acceptable range of variance ratio is set between 0.5 and 1.5, it is considered that the actual signal variance and the theoretical variance are consistent; the acceptable range of variance ratio is a tolerance interval set based on business experience.

[0106] The variance ratio - 1 calculates the absolute mutual error between the actual variance and the theoretical variance; the closer the variance ratio is to 1, the smaller this error is.

[0107] The similarity score is calculated as 1 - |variance ratio - 1|. The closer the score is to 1, the better the variance consistency.

[0108] Third step: Quantile matching test

[0109] Key quantiles of 5%, 25%, 50%, 75%, and 95% were selected. The matching degree between the actual signal quantiles and the theoretical quantiles was compared, the average quantile error was calculated, and the similarity score was obtained.

[0110] Similarity = 1 - Average Error

[0111] The above formula is used to evaluate the consistency between the actual signal distribution shape and the theoretical signal distribution;

[0112] The average error refers to the average of the absolute errors calculated at selected key points at 5%, 25%, 50%, 75%, and 95%, respectively, by subtracting the theoretical signal quantile from the actual signal quantile.

[0113] Subtract the average error to obtain the similarity score. The higher the score, the more similar the distribution shape. Set an error threshold to ensure the consistency of the distribution shape.

[0114] Step 6.2: Based on the results of the triple test, a weighted fusion algorithm is used to calculate the comprehensive similarity score, assigning 40% weight to the KS test, 35% weight to the variance test, and 25% weight to the quantile test to reflect the differences in importance of each test.

[0115] Step 6.3: Set the detection threshold to 0.7. When the overall similarity reaches or exceeds the threshold, the watermark is determined to exist. The confidence level is divided according to the score range: above 0.8 is high confidence, 0.6 to 0.8 is medium confidence, and below 0.6 is low confidence.

[0116] As a specific example, the watermark detection using the triple distribution similarity test mechanism described in step 6.1 involves designing a partial data comparison detection mechanism: when some original parameter data or suspected data is missing, signal extraction and subsequent triple distribution test are performed only based on the existing data points.

[0117] The system does not require the datasets to be 100% complete for comparison, but rather performs calculations based on existing data. When the data missing rate is less than 10%, the detection accuracy based on partial data can still remain above 90%, which demonstrates the robustness of the method to common data loss problems in real-world environments.

[0118] As a specific example, the detection threshold described in step 6.3 is dynamically adjusted based on data quality and the detection environment:

[0119] Standard thresholds are used when the data is complete, the sequence is stable, the computing power is high and the latency is low, and the data quality is high and the environment is stable.

[0120] In cases where the data missing rate is greater than 10%, the sequence fluctuates severely, computing power is limited, and the data quality is low due to high latency and complex environment, the threshold requirements can be relaxed to balance detection sensitivity and false alarm rate.

[0121] As a specific example, the generation of the detection technology report in step 7 is as follows:

[0122] Each test generates a detailed technical report, including signal statistical analysis, detailed results of each test item, comprehensive similarity score, watermark existence judgment and confidence level assessment; the report adopts a standardized format and has legal evidentiary effect.

[0123] The detection report serves as the final proof and closure loop for realizing the value of this method. It transforms the technical "watermark detection" results into legally and technically persuasive evidence. This report directly proves that the data is "encrypted and protected" by this method, enabling it to trace the source of data breaches, determine copyright ownership, and protect rights in disputes. Thus, "encryption protection" is no longer confined to the theoretical level but possesses practical application value and deterrent power.

[0124] Example

[0125] This embodiment uses 10,000 randomly generated photovoltaic power generation data points as an example for simulation experiments. The data range is (100.00, 999.99), the privacy budget ε=0.8, the confidence level is 99%, the maximum signal threshold is 5, and the signal mechanism is Laplace. The simulation equipment is:

[0126] CPU: Intel U7 265K; Clock speed: 5.2GHz;

[0127] Graphics card: Intel B580; Core: 2850MHz;

[0128] Memory: LEXAR THOR; Frequency: 6400 MHz;

[0129] SSD: WD SN5000; Write speed: 5100mb / s;

[0130] (1) Comparison of data distribution

[0131] The total processing time for adding a signal to 10,000 data points is approximately 0.0015 seconds. The data distribution before and after noise addition is as follows: Figure 2 As shown, the data distribution before and after watermarking shows slight skew at individual points, but remains consistent overall, balancing privacy and data usability.

[0132] (2) Data quality comparison

[0133] Comparison of data quality before and after watermarking, for example Figure 3 As shown, the data with added noise has up to 99% consistency with the original data in terms of mean, standard deviation, and range, which mathematically proves the usability of the data.

[0134] In summary, this invention provides a method for encrypting and protecting photovoltaic power generation data based on local differential privacy. This method achieves privacy protection at the data source, namely each photovoltaic site, preventing the leakage of raw data. The processed data retains the core characteristics of photovoltaic data, supporting accurate power generation forecasting and grid dispatching. Imperceptible but detectable identification information is embedded, enabling data traceability and copyright protection. Strict mathematical privacy guarantees are provided, resisting various privacy attacks and ensuring high security. Adapting to the high precision and temporal characteristics of photovoltaic power generation data, this invention provides key technical support for building a safe, efficient, and intelligent new power system.

[0135] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for encrypting and protecting photovoltaic power generation data based on local differential privacy, characterized in that, Includes the following steps: Step 1: Establish a privacy parameter configuration mechanism; Step 2: Generate a random signal using the Laplace mechanism; Step 3: Embed the signal as an invisible watermark into the original data at the data source to obtain protected data; Step 4: The system automatically records key parameters during the watermark embedding process, forming a complete embedding evidence chain; Step 5: The detection end calculates the actual signal sequence by comparing the suspected data with the original reference data, and performs integrity verification and outlier detection on the extracted signal sequence; Step 6: Use the triple distribution similarity test mechanism to detect the watermark and obtain the comprehensive similarity score and watermark confidence level; Step 7: Generate a testing technical report and complete the encryption protection of photovoltaic power generation data.

2. The photovoltaic power generation data encryption and protection method based on local differential privacy according to claim 1, characterized in that, The establishment of the privacy parameter configuration mechanism described in step 1 is as follows: Based on the sensitivity characteristics of photovoltaic data and the accuracy requirements of business operations, the optimal privacy parameters are dynamically calculated. The system presets a privacy budget of ε=0.8 and a data sensitivity of Δf=1.

0. Through the precise calculation of the Laplace signal scale parameter λ=Δf / ε, it ensures that data utility is maintained while providing strict privacy protection.

3. The photovoltaic power generation data encryption and protection method based on local differential privacy according to claim 2, characterized in that, Step 2, which involves generating a random signal using the Laplace mechanism, is detailed below: Step 2.1: Generate a random signal using the Laplace mechanism. The signal follows a Laplace distribution Laplace(0,λ). Calculate the theoretical quantiles and dynamically adjust the signal range to ensure that the absolute value of the signal does not exceed the set value at a 99% confidence level. Step 2.2: The system monitors the statistical characteristics of the generated signal in real time. When the theoretical signal limit exceeds the preset threshold, the signal truncation function is automatically activated to cut the signal value that exceeds the range to the allowable range. At the same time, the truncation ratio is recorded for quality assessment.

4. The photovoltaic power generation data encryption and protection method based on local differential privacy according to claim 1, characterized in that, Step 3 involves embedding the signal as an invisible watermark into the original data at the data source to obtain protected data, as detailed below: For the power generation data of each photovoltaic site, the system generates a corresponding Laplace signal, which is then embedded as an invisible watermark into the original data to obtain protected data. Protected data = raw data + controlled signals The raw data refers to the actual power generation data collected by the photovoltaic site; the controlled signal refers to the random signal generated by the Laplace mechanism and controlled by range; the protected data refers to the result after the signal is embedded, which is both privacy-preserving data that satisfies local differential privacy and copyright identification data with embedded invisible watermark. The embedding process ensures that the temporal and statistical characteristics of the data are preserved, and the correlation coefficient between the watermarked data and the original data is maintained above 0.

98.

5. The photovoltaic power generation data encryption and protection method based on local differential privacy according to claim 1, characterized in that, The key parameters in the watermark embedding process described in step 4 include privacy budget, signal scale, actual signal statistics, and embedding timestamp.

6. The photovoltaic power generation data encryption and protection method based on local differential privacy according to claim 1, characterized in that, The formula for calculating the actual signal sequence mentioned in step 5 is as follows: Actual signal = Suspected data - Original data The above formula describes how to extract possible watermarks, i.e., signal sequences, from the data to be verified in the detection segment; The suspected data refers to data that is discovered during circulation or use and whose source and copyright need to be verified. The raw data refers to the real data used as a comparison benchmark, which has not been processed in any way; The actual signal is the signal sequence extracted through differential calculation; if the suspected data is the protected data obtained in step 3, then the extracted actual signal and the controlled signal during embedding are consistent in statistical distribution.

7. The photovoltaic power generation data encryption and protection method based on local differential privacy according to claim 1, characterized in that, Step 6 describes the use of a triple distribution similarity test mechanism for watermark detection, yielding a comprehensive similarity score and watermark confidence level, as detailed below: Step 6.1: Use the triple distribution similarity test mechanism to detect watermarks. First step: Consistency test of KS distribution The Kolmogorov-Smirnov test is used to compare the consistency between the actual signal distribution and the theoretical Laplace distribution. The KS statistic and the corresponding p-value are calculated, and the statistic is converted into a similarity score. Similarity = 1 - KS statistic The significance level was set at 1%, and a p-value greater than 0.01 was considered a consistent distribution. Second step: Variance consistency test By comparing the consistency between the actual signal variance and the theoretical variance, and calculating the variance ratio, a similarity assessment can be obtained. Similarity = 1 - |variance ratio - 1| The above formula is used to evaluate whether the dispersion of the actual signal is consistent with the expected theory; Variance ratio = Actual signal variance / Theoretical variance, where the theoretical variance of the Laplace mechanism is 2λ²; when the acceptable range of variance ratio is set between 0.5 and 1.5, it is considered that the actual signal variance and the theoretical variance are consistent; the acceptable range of variance ratio is a tolerance interval set based on business experience. The variance ratio - 1 calculates the absolute mutual error between the actual variance and the theoretical variance; the closer the variance ratio is to 1, the smaller this error is. The similarity score is calculated as 1 - |variance ratio - 1|. The closer the score is to 1, the better the variance consistency. Third step: Quantile matching test Key quantiles of 5%, 25%, 50%, 75%, and 95% were selected. The matching degree between the actual signal quantiles and the theoretical quantiles was compared, the average quantile error was calculated, and the similarity score was obtained. Similarity = 1 - Average Error The above formula is used to evaluate the consistency between the actual signal distribution shape and the theoretical signal distribution; The average error refers to the average of the absolute errors calculated at selected key points at 5%, 25%, 50%, 75%, and 95%, respectively, by subtracting the theoretical signal quantile from the actual signal quantile. Subtract the average error to obtain the similarity score. The higher the score, the more similar the distribution shape. Set an error threshold to ensure the consistency of the distribution shape. Step 6.2: Based on the results of the triple test, a weighted fusion algorithm is used to calculate the comprehensive similarity score, assigning 40% weight to the KS test, 35% weight to the variance test, and 25% weight to the quantile test to reflect the differences in importance of each test. Step 6.3: Set the detection threshold to 0.

7. When the overall similarity reaches or exceeds the threshold, the watermark is determined to exist. The confidence level is divided according to the score range: above 0.8 is high confidence, 0.6 to 0.8 is medium confidence, and below 0.6 is low confidence.

8. The photovoltaic power generation data encryption and protection method based on local differential privacy according to claim 7, characterized in that, The watermark detection using the triple distribution similarity test mechanism described in step 6.1 involves designing a partial data comparison detection mechanism: when some original parameter data or suspected data is missing, signal extraction and subsequent triple distribution test are performed only based on the existing data points. The system does not require the dataset for comparison to be 100% complete, but rather performs calculations based on existing data; when the data missing rate is within 10%, the detection accuracy based on partial data can still remain above 90%.

9. The photovoltaic power generation data encryption and protection method based on local differential privacy according to claim 7, characterized in that, The detection threshold mentioned in step 6.3 is dynamically adjusted based on data quality and the detection environment. Standard thresholds are used when the data is complete, the sequence is stable, the computing power is high and the latency is low, and the data quality is high and the environment is stable. In cases where the data missing rate is greater than 10%, the sequence fluctuates severely, computing power is limited, and the data quality is low due to high latency and complex environment, the threshold requirements can be relaxed to balance detection sensitivity and false alarm rate.

10. The photovoltaic power generation data encryption and protection method based on local differential privacy according to claim 1, characterized in that, Step 7, generating the detection technical report, is as follows: Each test generates a detailed technical report, including signal statistical analysis, detailed results of each test item, comprehensive similarity score, watermark existence judgment and confidence level assessment; the report adopts a standardized format and has legal evidentiary effect.

Images