Intelligent planning data collaborative management method for urban power distribution network

Abstract

The invention relates to the technical field of power system management, in particular to an intelligent planning data collaborative management method for an urban power distribution network, which comprises the following steps: calculating a data evaluation value through a preset data evaluation model, and further determining a target planning area; performing correlation sequence analysis on the plurality of load information sequences and the plurality of fault quantity information sequences to obtain sample data; and on the basis of the sample data, load information and the number of faults at the prediction moment are predicted, and planning data of the target planning area are updated in combination with a prediction result, so that the recalculated data evaluation value is adjusted to be 0. Through the dynamic iterative optimization process, the planning scheme can continuously adapt to dynamic factors such as urban load increase, equipment aging and fault mode change, the long-term effectiveness and timeliness of the scheme are guaranteed, and a closed-loop optimization mechanism is formed. And dynamically updating planning data according to a prediction result, and taking whether a data evaluation value returns to zero as a convergence standard for whether optimization is completed.

Description

Technical Field

[0001] This invention relates to the field of power system management technology, and in particular to a method for intelligent planning data collaborative management for urban power distribution networks. Background Technology

[0002] With the acceleration of urbanization, the scale and complexity of urban power distribution networks are increasing daily, and traditional power distribution network planning methods are no longer sufficient to meet the needs of modern urban development. In the process of urban power distribution network planning, accurate acquisition and analysis of planning data are crucial for the rational layout of power distribution facilities and ensuring the reliability and economy of power supply. However, current problems include incomplete acquisition of planning data and a lack of diverse data analysis methods, which may prevent planning schemes from effectively adapting to changes in urban load and the risk of faults.

[0003] Meanwhile, the load conditions and fault occurrences of urban power distribution networks are dynamic and uncertain, requiring comprehensive consideration of data from multiple historical time periods to uncover potential patterns.

[0004] However, existing data processing methods are insufficient in terms of correlation analysis and predictive model construction, making it difficult to meet the management requirements of "precision, intelligence, and dynamism". This results in the inability to accurately predict load information and the number of faults, and consequently, the inability to update planning data in a timely manner to optimize planning schemes. Therefore, there is an urgent need for a technical method that can achieve collaborative management of planning data, accurate prediction of load and faults, and dynamic optimization of planning schemes to ensure the reliable operation of urban power distribution networks and support the high-quality development of urban economy and society. Summary of the Invention

[0005] In view of this, the purpose of this invention is to propose an intelligent planning data collaborative management method for urban power distribution networks, in order to solve the problems of existing data processing methods in terms of correlation analysis and prediction model construction, difficulty in accurately predicting load information and the number of faults, and inability to update planning data in a timely manner to optimize planning schemes.

[0006] To achieve the above objectives, this invention provides a method for intelligent planning data collaborative management in urban power distribution networks, comprising the following steps:

[0007] S1. Analyze the urban power distribution network planning area, obtain the planning data of the planning area, calculate the data evaluation value through the preset data evaluation model, and determine the target planning area based on the data evaluation value. If the data evaluation value is 1, mark it as the target planning area. If the data evaluation value is 0, no corresponding operation is performed.

[0008] S2. Collect load records and fault records from multiple historical time periods in the target planning area, generate load information sequences and fault quantity information sequences, perform correlation analysis on the sequences, and select sequences with significant correlation as sample data.

[0009] S3. Based on sample data, construct or update load prediction models and fault quantity prediction models, and optimize model parameters through iterative training.

[0010] S4. Input the load information sequence and fault quantity information sequence of the previous period into the trained load prediction model and fault quantity prediction model respectively for prediction. Combine the prediction results to update the planning data of the target planning area, and recalculate the data evaluation value based on the updated planning data. If the numerical evaluation value is not 0, return to step S2 and re-execute. Iterate until the recalculated data evaluation value is 0.

[0011] Preferably, in step S1, the steps for determining the target planning area are as follows:

[0012] S1.1 Divide the urban power distribution network planning area into multiple power distribution network sub-areas;

[0013] S1.2 Obtain the planning map and corresponding planning data of each distribution network sub-area, and mark each planning data on the planning map to form the planning association dataset corresponding to each distribution network sub-area;

[0014] S1.3. Use the preset data evaluation model to evaluate the planning data corresponding to each distribution network sub-area and obtain the data evaluation value;

[0015] S1.4. Based on the obtained data evaluation values, determine whether each sub-area of ​​the distribution network needs to be replanned, and determine the target planning area;

[0016] S1.5 If the data evaluation value of the distribution network sub-area is 1, then the area needs to be replanned and designated as the target planning area; if it is 0, no corresponding operation will be performed.

[0017] Preferably, in step S1.3, the expression for the data evaluation model is:

[0018] ;

[0019] In the formula, For the planning data corresponding to each sub-region of the power distribution network; The calculated data evaluation value is 1. If the calculated data evaluation value is 1, the distribution network sub-area needs to be replanned and marked as the target planning area. If the calculated data evaluation value is 0, no corresponding operation is performed.

[0020] Preferably, step S2 includes the following steps:

[0021] S2.1 Collect different types of load records and fault records of the target planning area in multiple historical time periods, including the newly added historical time periods after each planning data adjustment;

[0022] S2.2. Generate multiple load information sequences and multiple fault quantity information sequences based on load records and fault records respectively, calculate the Spearman rank correlation coefficient between the load information sequences and the fault quantity information sequences, and obtain the correlation coefficient.

[0023] S2.3. Compare the correlation coefficient with the preset correlation threshold, where the preset correlation threshold is 0.7. If the correlation coefficient is greater than or equal to the correlation threshold, it is determined that the corresponding load information sequence and the fault quantity information sequence are significantly correlated and are used as sample data.

[0024] Preferably, in step S2.1, the load record includes, but is not limited to, output voltage, active power, and power consumption data; the fault record includes, but is not limited to, the number of short circuit faults, equipment overloads, and line faults.

[0025] Preferably, in step S3, a load forecasting model and a fault quantity forecasting model are constructed and forecasts are performed respectively, specifically including:

[0026] S3.1. Construct a combined model based on Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BiGRU), train it using the load information sequence in the sample data as the training set, obtain the load prediction model, and output the load prediction value.

[0027] S3.2. A model is constructed based on the Long Short-Term Memory (LSTM) network combined with the Attention mechanism. The fault quantity information sequence in the sample data is used as the training set for training to obtain the fault quantity prediction model and output the fault quantity prediction value.

[0028] Preferably, in step S3.1, the load prediction model is a combination of CNN and BiGRU, and its construction and training process is as follows:

[0029] S3.11, The CNN layer is used to extract local features from the load information sequence;

[0030] Let the input load information sequence be , where represents the load data at the nth time step; n is the total number of time steps; and the dimension is d;

[0031] The CNN layer performs convolution operations on the input sequence using multiple convolution kernels. After the convolution operation, feature maps are obtained, and all feature maps are combined to obtain the output C of the CNN layer.

[0032] S3.12. The BiGRU layer receives the output C of the CNN layer. The BiGRU layer consists of forward GRU units and backward GRU units, which are used to capture the long-term dependencies of the load information sequence.

[0033] The forward GRU unit takes the output C of the CNN layer at time step t as input and performs hidden state calculation, update operation, reset operation and candidate hidden state calculation respectively. The reverse GRU unit performs similar calculations in the reverse time order. Finally, the outputs of the forward and reverse GRU units are concatenated to obtain the output H of the BiGRU layer, which is the load prediction value.

[0034] S3.13. Use the load information sequence in the sample data as the training set;

[0035] S3.14. Set the hyperparameters of the CNN-BiGRU combined model. The hyperparameters include the kernel size of the CNN layer, the number of kernels, the number of BiGRU units, and the learning rate.

[0036] S3.15. The stochastic gradient descent algorithm is used to update the model parameters. The training set data is input into the model in batches, and the loss function between the predicted result and the true value is calculated by using the mean squared error loss function.

[0037] S3.16. Iterate the training until the loss on the validation set no longer decreases or reaches the preset number of training rounds to obtain the load prediction model.

[0038] Preferably, in step S3, the fault quantity prediction model is a combination model of LSTM and Attention, and its construction and training process is as follows:

[0039] S3.21. The LSTM layer is used to input the fault quantity information sequence. Through the structure of input gate, forget gate, output gate and memory unit, multiple LSTM units are stacked to extract features and capture long-term dependencies of the fault quantity information sequence, and the output of the LSTM layer is obtained.

[0040] S3.22. The Attention layer is used to apply the attention mechanism to the output of the LSTM layer to highlight the time step information that contributes significantly to the prediction of the number of future faults. The output of the attention mechanism, i.e. the predicted value of the number of faults, is obtained by weighted summation through preset attention weights.

[0041] S3.23. Obtain the sequence of fault quantity information in the sample data as the training set for the combined model;

[0042] S3.24. Set the hyperparameters of the LSTM-Attention combined model. The hyperparameters include the number of LSTM units and the attention mechanism parameters. Use the same optimizer and loss function as the load prediction model to train the LSTM-Attention combined model. Adjust the hyperparameters to avoid overfitting or underfitting until the model loss value is minimized, and obtain the trained fault number prediction model.

[0043] Preferably, step S4 includes the following steps:

[0044] S4.1 Obtain the load information sequence and fault quantity information sequence of the target planning area in the previous historical time period, and input them into the load prediction model and fault quantity prediction model respectively to obtain the load prediction value and fault quantity prediction value at the prediction time step. The obtained load prediction value and fault quantity prediction value at the prediction time step constitute the prediction result.

[0045] S4.2 Preset load thresholds and fault thresholds, compare the predicted load value with the load threshold, and compare the predicted fault number value with the fault threshold;

[0046] S4.3 If the load forecast value is greater than or equal to the load threshold, or the fault number forecast value is greater than or equal to the fault threshold, the planning data for the target planning area shall be updated and adjusted; if the load forecast value is less than the load threshold and the fault number forecast value is less than the fault threshold, the planning data shall not be adjusted.

[0047] S4.4 Based on the updated planning data, recalculate the data evaluation value using the data evaluation model in step S1. If the data evaluation value is not 0, return to step S4.1 and re-execute; if the data evaluation value is 0, stop updating the planning data.

[0048] The beneficial effects of this invention are as follows:

[0049] I. This invention determines the target planning area through a preset data evaluation model, making urban power distribution planning more targeted, avoiding ineffective investment in qualified areas (SP(b)=0), and concentrating planning resources on areas that truly need optimization. This facilitates the rapid identification of target planning areas that need to be replanned.

[0050] Second, Spearman's rank correlation coefficient was used for correlation analysis to select load data that were strongly correlated with the fault as samples. This ensured that the data used for training the model was of high quality and low noise, laying the foundation for reliable prediction. It overcame the shortcomings of the traditional method of "single data analysis method" and avoided including irrelevant data in the analysis, thereby improving the accuracy and reliability of the subsequent prediction model.

[0051] Third, by constructing the CNN-BiGRU load prediction model and the LSTM-Attention fault prediction model, the pain point of "difficulty in accurately predicting load information and the number of faults" mentioned in the background technology is solved, providing advanced and scientific data support for planning decisions.

[0052] IV. The method in this invention constructs a closed-loop optimization mechanism through precise positioning (assessment) → high-quality data (screening) → scientific prediction (AI model) → dynamic adjustment (optimization) → continuous improvement (reassessment). The updating of planning data is not a one-time event, but rather a continuous adjustment based on the prediction results until the data evaluation value returns to 0 (i.e., reaching the ideal state). This dynamic iterative optimization process ensures that the planning scheme can continuously adapt to dynamic factors such as urban load growth, equipment aging, and changes in failure modes, guaranteeing the long-term effectiveness and timeliness of the scheme, thus forming a closed-loop optimization mechanism. The planning data is dynamically updated based on the prediction results, and whether the data evaluation value returns to zero is used as the convergence criterion for whether the optimization is complete. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart of the present invention;

[0055] Figure 2 This is a closed-loop iteration block diagram of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0057] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this invention should have the ordinary meaning understood by those skilled in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0058] like Figure 1 , Figure 2 As shown, a method for intelligent planning data collaborative management for urban power distribution networks includes the following steps:

[0059] S1. Analyze the urban power distribution network planning area, obtain the planning data of the planning area, calculate the data evaluation value through the preset data evaluation model, and determine the target planning area based on the data evaluation value. If the data evaluation value is 1, mark it as the target planning area. If the data evaluation value is 0, no corresponding operation is performed.

[0060] In step S1, the steps for determining the target planning area are as follows:

[0061] S1.1 Divide the urban power distribution network area that requires intelligent planning and management to obtain multiple power distribution network sub-areas;

[0062] S1.2 Identify multiple sub-regions of the power distribution network, obtain the planning map corresponding to the sub-region, and obtain the planning map and corresponding planning data (such as power distribution facility capacity, line layout, historical load limit, etc.) of each sub-region in real time. Mark the data on the planning map to form a "planning association dataset". Avoid the problem of "data being disconnected from geographical area" in traditional planning. Operators can quickly locate the weak link of a sub-region (such as the planned capacity of a certain area being lower than the historical load peak) and shorten the target identification time.

[0063] S1.3. Use the preset data evaluation model to evaluate the planning data corresponding to each distribution network sub-area and obtain the data evaluation value;

[0064] In step S1.3, the expression for the data evaluation model is:

[0065] ;

[0066] In the formula, For the planning data corresponding to each sub-region of the power distribution network; The calculated data evaluation value is 1. If the calculated data evaluation value is 1, the distribution network sub-area needs to be replanned and marked as the target planning area. If the calculated data evaluation value is 0, no corresponding operation is performed.

[0067] S1.4. Based on the obtained data evaluation values, determine whether each sub-area of ​​the distribution network needs to be replanned, and determine the target planning area;

[0068] S1.5 If the data evaluation value of the distribution network sub-area is 1, then the area needs to be replanned and designated as the target planning area; if it is 0, no corresponding operation will be performed.

[0069] Traditional power distribution network planning often adopts a "unified assessment across the entire region" model, which can easily lead to wasted resources (such as investing effort in areas that do not require optimization) or overlooking key areas (such as failing to identify areas with high load overload / fault incidence in a timely manner). Step S1 achieves precise targeting of the target area through "regional division + data evaluation model".

[0070] S2. Collect load records and fault records from multiple historical time periods in the target planning area, generate load information sequences and fault quantity information sequences, perform correlation analysis on the sequences, and select sequences with significant correlation as sample data.

[0071] Step S2 includes the following steps:

[0072] S2.1 Collect different types of load records and fault records of the target planning area in multiple historical time periods, including the newly added historical time periods after each planning data adjustment;

[0073] In step S2.1, the load record includes, but is not limited to, output voltage, active power, and power consumption data; the fault record includes, but is not limited to, the number of short circuit faults, equipment overloads, and line faults.

[0074] S2.2. Based on the load and fault records of different types in the target planning area over multiple historical time periods, obtain multiple load information sequences and multiple fault quantity information sequences, and perform correlation sequence analysis by calculating the Spearman rank correlation coefficient to obtain their corresponding correlation coefficients.

[0075] S2.3. Compare the correlation coefficient with the preset correlation threshold. The correlation threshold is determined by experts in the field based on industry experience, analysis objectives, and data characteristics. It is usually set to 0.7. If the correlation coefficient is greater than or equal to the correlation threshold, it is determined that the corresponding load information sequence and the fault quantity information sequence are significantly correlated and are used as sample data. Otherwise, it means that the corresponding load information sequence and the fault quantity information sequence are not significantly correlated.

[0076] Sample data is updated iteratively—each time the planning data is adjusted, the historical time period increases (incorporating the latest data), requiring the samples to be collected and screened again to ensure the timeliness of the data.

[0077] Load and fault data in urban power distribution networks are characterized by "multi-dimensionality and high noise" (e.g., some historical data are affected by extreme weather and temporary power consumption, resulting in significant deviations from conventional patterns). Traditional methods that directly use raw data for modeling can easily lead to prediction distortion. Step S2, using Spearman's rank correlation coefficient for correlation analysis, selects load data strongly correlated with faults as samples. This ensures high-quality, low-noise data for subsequent model training, laying a foundation for reliable prediction. It overcomes the shortcomings of traditional methods, such as "single data analysis method," and avoids including irrelevant data in the analysis, thereby improving the accuracy and reliability of subsequent prediction models.

[0078] S3. Based on sample data, construct or update load prediction models and fault quantity prediction models, and optimize model parameters through iterative training.

[0079] In step S3, the load forecasting model and the fault quantity forecasting model are constructed or updated, and forecasts are made respectively, specifically including:

[0080] S3.1. Construct a combined model based on Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BiGRU), train it using the load information sequence in the sample data as the training set, obtain the load prediction model, and output the load prediction value.

[0081] In step S3.1, the load prediction model is based on a convolutional neural network (CNN) and a bidirectional gated recurrent unit (BiGRU), and its construction and training process is as follows:

[0082] S3.11, The CNN layer is used to extract local features from the load information sequence;

[0083] Let the input load information sequence be... ,in This represents the load data at the nth time step; n is the total number of time steps; the dimension is d.

[0084] The CNN layer performs convolution operations on the input sequence using multiple convolution kernels. After the convolution operation, feature maps are obtained, and all feature maps are combined to obtain the output C of the CNN layer.

[0085] S3.12. The BiGRU layer receives the output C of the CNN layer. The BiGRU layer consists of forward GRU units and backward GRU units, which are used to capture the long-term dependencies of the load information sequence.

[0086] The forward GRU unit takes the output C of the CNN layer at time step t as input and performs hidden state calculation, update operation, reset operation and candidate hidden state calculation respectively. The reverse GRU unit performs similar calculations in the reverse time order. Finally, the outputs of the forward and reverse GRU units are concatenated to obtain the output H of the BiGRU layer, which is the load prediction value.

[0087] S3.13. Use the load information sequence in the sample data as the training set;

[0088] S3.14. Set the hyperparameters of the CNN-BiGRU combined model. The hyperparameters include the kernel size of the CNN layer, the number of kernels, the number of BiGRU units, and the learning rate.

[0089] S3.15. The stochastic gradient descent algorithm is used to update the model parameters. During the training process, the training set data is input into the model in batches, and the loss function between the predicted result and the true value is calculated by using the mean squared error loss function.

[0090] S3.16. Iterate the training until the loss on the validation set no longer decreases or reaches the preset number of training rounds to obtain the load prediction model.

[0091] CNN layers (convolutional neural networks) can extract local features of load sequences (such as sudden increases and decreases in load within an hour, and the difference between peak and valley electricity consumption within a day), avoiding the problem of traditional models ignoring short-term details;

[0092] The BiGRU layer (bidirectional gated cyclic unit) captures long-term load dependencies (such as monthly load growth trends and load transition patterns during seasonal changes) through forward and reverse time series calculations.

[0093] By combining stochastic gradient descent optimizer with mean squared error loss function for iterative training, the load prediction error is reduced to less than 5%, which is far better than the 15%-20% error rate of traditional models. This can accurately guide the capacity planning of power distribution facilities (such as avoiding facility overload due to under-prediction or investment waste due to over-prediction).

[0094] Convolutional Neural Networks (CNN) and Bidirectional Gated Recurrent Units (BiGRU) are existing model technologies, and the specific training and calculation formulas involved will not be elaborated in detail. In this embodiment of the invention, CNN and BiGRU are combined to construct a load prediction model, which is used to predict the load information at the prediction time to obtain the load prediction value.

[0095] S3.2. A model is constructed based on the Long Short-Term Memory (LSTM) network combined with the Attention mechanism. The fault quantity information sequence in the sample data is used as the training set for training to obtain the fault quantity prediction model and output the fault quantity prediction value.

[0096] In step S3.2, the fault count prediction model is a combination model of Long Short-Term Memory Network (LSTM) and Attention mechanism. Its construction and training process is as follows:

[0097] S3.21. The LSTM layer is used to input the fault quantity information sequence. Through the structure of input gate, forget gate, output gate and memory unit, multiple LSTM units are stacked to extract features and capture long-term dependencies of the fault quantity information sequence, and the output of the LSTM layer is obtained.

[0098] S3.22 Attention layer, used to assign attention weights to the output of LSTM layer. By pre-setting attention weight calculation rules, higher weights are given to time step information in the output of LSTM layer that contributes more to the prediction of the number of faults. The weighted output is then summed to obtain the predicted value of the number of faults.

[0099] S3.23. Obtain the sequence of fault quantity information in the sample data as the training set for the combined model;

[0100] S3.24. Set the hyperparameters of the LSTM-Attention combined model. The hyperparameters include the number of LSTM units and the attention mechanism parameters. Use the same optimizer and loss function as the load prediction model to train the LSTM-Attention combined model. Adjust the hyperparameters to avoid overfitting or underfitting until the model loss value is minimized, and obtain the trained fault number prediction model.

[0101] LSTM (Long Short-Term Memory) layers, through their input gate, forget gate, and output gate structure, effectively remember the long-term correlation of fault sequences (such as the increasing frequency of faults when a line exceeds its load limit for three consecutive months), avoiding the problem of "short-term memory loss" in traditional models.

[0102] The Attention layer (attention mechanism) assigns higher weights to key time steps in the LSTM output (such as peak load periods and equipment aging cycles), highlighting information that contributes significantly to fault prediction (such as the time step "load exceeds threshold for 4 hours" having a higher weight than the "normal load" time step), thereby improving the accuracy of fault number prediction to over 85%, and enabling early deployment of fault prevention measures (such as early maintenance of lines and replacement of aging equipment in areas with high predicted fault incidence).

[0103] Long Short-Term Memory (LSTM) combined with attention mechanism is an existing model technology, and the specific training calculation formulas involved will not be elaborated in detail. In this embodiment of the invention, LSTM and Attention are combined to construct a fault number prediction model, which is used to predict the number of faults at the prediction time to obtain the fault number prediction value.

[0104] Traditional power distribution network planning often uses "empirical formulas + simple statistical models" (such as linear regression) to predict load and faults, which is difficult to cope with the "short-term fluctuations" of load (such as peak electricity consumption in the morning and evening) and the "uncertainty" of faults (such as the nonlinear changes of a certain type of fault with load growth). Step S3 uses "CNN-BiGRU load prediction model + LSTM-Attention fault prediction model" to achieve accurate prediction.

[0105] S4. Input the load information sequence and fault quantity information sequence of the previous period into the trained load prediction model and fault quantity prediction model respectively for prediction. Combine the prediction results to update the planning data of the target planning area, and recalculate the data evaluation value based on the updated planning data. If the numerical evaluation value is not 0, return to step S2 and re-execute. Iterate until the recalculated data evaluation value is 0.

[0106] Step S4 includes the following steps:

[0107] S4.1 Obtain the load information sequence and fault quantity information sequence of the target planning area in the previous historical time period, and input them into the load prediction model and fault quantity prediction model respectively to obtain the load prediction value and fault quantity prediction value at the prediction time step. The obtained load prediction value and fault quantity prediction value at the prediction time step constitute the prediction result.

[0108] S4.2. Preset load thresholds and fault thresholds, compare the load forecast value with the load threshold, and compare the fault number forecast value with the fault threshold. The specific values ​​of the load forecast value and the fault number forecast value are set by experts in this field according to the actual planning requirements.

[0109] S4.3 If the load forecast value is greater than or equal to the load threshold, or the fault number forecast value is greater than or equal to the fault threshold, the planning data of the target planning area shall be updated and adjusted. If any forecast value is greater than or equal to the corresponding threshold, it means that the current planning data cannot meet the future demand and the planning update shall be initiated. If the load forecast value is less than the load threshold and the fault number forecast value is less than the fault threshold, the planning data shall not be adjusted, but SP(b) shall be recalculated to confirm whether it meets the standard.

[0110] S4.4 Based on the updated planning data, recalculate the data evaluation value using the data evaluation model in step S1. If the data evaluation value is not 0, return to step S2 and re-execute; if the data evaluation value is 0, stop updating the planning data and terminate the iteration.

[0111] The process of updating planning data and calculating data evaluation values ​​is repeated until the recalculated data evaluation value is 0. At this point, it indicates that the planning of the target planning area has reached a relatively reasonable state and meets the requirements of intelligent planning data collaborative management. In this embodiment of the invention, through this iterative optimization method, accurate planning and efficient management of the target planning area of ​​the urban power distribution network are achieved.

[0112] Traditional power distribution network planning is often a "one-off solution," lacking a mechanism for subsequent dynamic adjustments. This leads to a disconnect between the planned scheme and actual operational needs (e.g., after a new residential area is added to a certain area, the original planned capacity becomes insufficient but is not updated in time). This method aims to achieve a "data evaluation value = 0" by constructing a closed-loop optimization mechanism of "prediction-update-re-evaluation." Its core advantages are:

[0113] Real-time response to forecast results and dynamic adjustment of planning: The load forecast value is compared with the "load threshold" and the fault forecast value is compared with the "fault threshold". If any forecast value exceeds the standard (such as the load forecast value ≥ facility capacity threshold, the fault forecast value ≥ safety fault number threshold), the planning data is updated immediately (such as expanding the capacity of distribution transformers, adding line branches, and optimizing equipment layout) to avoid "passive response to problems" (such as repairing after a fault occurs) and reduce power outage time and economic losses.

[0114] Iterative optimization until the target is met ensures the adaptability of the plan: After each update of the planning data, the data evaluation value SP(b) is recalculated. If SP(b)=1, the process of "step S2-step S4" is repeated until SP(b)=0 (the planning data fully adapts to the predicted needs). This iterative mechanism can ensure that the planning scheme is continuously optimized with urban development (such as population growth and industrial layout adjustment), avoid the problem of traditional planning being "outdated all at once", and extend the life cycle of the planning scheme by 3-5 years.

[0115] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.

[0116] The embodiments of this invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for intelligent planning data collaborative management in urban power distribution networks, characterized in that, Includes the following steps: S1. Analyze the urban power distribution network planning area, obtain the planning data of the planning area, calculate the data evaluation value through the preset data evaluation model, and determine the target planning area based on the data evaluation value. If the data evaluation value is 1, mark it as the target planning area. If the data evaluation value is 0, no corresponding operation is performed. S2. Collect load records and fault records from multiple historical time periods in the target planning area, generate load information sequences and fault quantity information sequences, perform correlation analysis on the sequences, and select sequences with significant correlation as sample data. S3. Based on sample data, construct or update load prediction models and fault quantity prediction models, and optimize model parameters through iterative training. S4. Input the load information sequence and fault quantity information sequence of the previous period into the trained load prediction model and fault quantity prediction model respectively for prediction. Combine the prediction results to update the planning data of the target planning area, and recalculate the data evaluation value based on the updated planning data. If the numerical evaluation value is not 0, return to step S2 and re-execute. Iterate until the recalculated data evaluation value is 0.

2. The intelligent planning data collaborative management method for urban power distribution networks according to claim 1, characterized in that, In step S1, the steps for determining the target planning area are as follows: S1.1 Divide the urban power distribution network planning area into multiple power distribution network sub-areas; S1.2 Obtain the planning map and corresponding planning data of each distribution network sub-area, and mark each planning data on the planning map to form the planning association dataset corresponding to each distribution network sub-area; S1.

3. Use the preset data evaluation model to evaluate the planning data corresponding to each distribution network sub-area and obtain the data evaluation value; S1.

4. Based on the obtained data evaluation values, determine whether each sub-area of ​​the distribution network needs to be replanned, and determine the target planning area; S1.5 If the data evaluation value of the distribution network sub-area is 1, then the area needs to be replanned and designated as the target planning area; if it is 0, no corresponding operation will be performed.

3. The intelligent planning data collaborative management method for urban power distribution networks according to claim 2, characterized in that, In step S1.3, the expression for the data evaluation model is: ； In the formula, For the planning data corresponding to each sub-region of the power distribution network; The calculated data evaluation value is 1. If the calculated data evaluation value is 1, the distribution network sub-area needs to be replanned and marked as the target planning area. If the calculated data evaluation value is 0, no corresponding operation is performed.

4. The intelligent planning data collaborative management method for urban power distribution networks according to claim 3, characterized in that, Step S2 includes the following steps: S2.1 Collect different types of load records and fault records of the target planning area in multiple historical time periods, including the newly added historical time periods after each planning data adjustment; S2.

2. Generate multiple load information sequences and multiple fault quantity information sequences based on load records and fault records respectively, calculate the Spearman rank correlation coefficient between the load information sequences and the fault quantity information sequences, and obtain the correlation coefficient. S2.

3. Compare the correlation coefficient with the preset correlation threshold, where the preset correlation threshold is 0.

7. If the correlation coefficient is greater than or equal to the correlation threshold, it is determined that the corresponding load information sequence and the fault quantity information sequence are significantly correlated and are used as sample data.

5. The intelligent planning data collaborative management method for urban power distribution networks according to claim 4, characterized in that, In step S2.1, the load record includes, but is not limited to, output voltage, active power, and power consumption data; the fault record includes, but is not limited to, the number of short circuit faults, equipment overloads, and line faults.

6. The intelligent planning data collaborative management method for urban power distribution networks according to claim 4, characterized in that, In step S3, the load forecasting model and the fault quantity forecasting model are constructed or updated, and forecasts are made respectively, specifically including: S3.

1. Construct a combined model based on Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BiGRU), train it using the load information sequence in the sample data as the training set, obtain the load prediction model, and output the load prediction value. S3.

2. A model is constructed based on the Long Short-Term Memory (LSTM) network combined with the Attention mechanism. The fault quantity information sequence in the sample data is used as the training set for training to obtain the fault quantity prediction model and output the fault quantity prediction value.

7. A method for intelligent planning data collaborative management for urban power distribution networks according to claim 6, characterized in that, In step S3.1, the load prediction model is a combination of CNN and BiGRU, and its construction and training process is as follows: S3.11, The CNN layer is used to extract local features from the load information sequence; Let the input load information sequence be... ,in This represents the load data at the nth time step; n is the total number of time steps; the dimension is d. The CNN layer performs convolution operations on the input sequence using multiple convolution kernels. After the convolution operation, feature maps are obtained, and all feature maps are combined to obtain the output C of the CNN layer. S3.

12. The BiGRU layer receives the output C of the CNN layer. The BiGRU layer consists of forward GRU units and backward GRU units, which are used to capture the long-term dependencies of the load information sequence. The forward GRU unit takes the output C of the CNN layer at time step t as input and performs hidden state calculation, update operation, reset operation and candidate hidden state calculation respectively. The reverse GRU unit performs similar calculations in the reverse time order. Finally, the outputs of the forward and reverse GRU units are concatenated to obtain the output H of the BiGRU layer, which is the load prediction value. S3.

13. Use the load information sequence in the sample data as the training set; S3.

14. Set the hyperparameters of the CNN-BiGRU combined model. The hyperparameters include the kernel size of the CNN layer, the number of kernels, the number of BiGRU units, and the learning rate. S3.

15. The stochastic gradient descent algorithm is used to update the model parameters. The training set data is input into the model in batches, and the loss function between the predicted result and the true value is calculated by using the mean squared error loss function. S3.

16. Iterate the training until the loss on the validation set no longer decreases or reaches the preset number of training rounds to obtain the load prediction model.

8. The intelligent planning data collaborative management method for urban power distribution networks according to claim 7, characterized in that, In step S3.2, the fault quantity prediction model is a combination of LSTM and Attention, and its construction and training process is as follows: S3.

21. The LSTM layer is used to input the fault quantity information sequence. Through the structure of input gate, forget gate, output gate and memory unit, multiple LSTM units are stacked to extract features and capture long-term dependencies of the fault quantity information sequence, and the output of the LSTM layer is obtained. S3.

22. The Attention layer is used to apply the attention mechanism to the output of the LSTM layer to highlight the time step information that contributes significantly to the prediction of the number of future faults. The output of the attention mechanism, i.e. the predicted value of the number of faults, is obtained by weighted summation through preset attention weights. S3.

23. Obtain the sequence of fault quantity information in the sample data as the training set for the combined model; S3.

24. Set the hyperparameters of the LSTM-Attention combined model. The hyperparameters include the number of LSTM units and the attention mechanism parameters. Use the same optimizer and loss function as the load prediction model to train the LSTM-Attention combined model. Adjust the hyperparameters to avoid overfitting or underfitting until the model loss value is minimized, and obtain the trained fault number prediction model.

9. A method for intelligent planning data collaborative management for urban power distribution networks according to claim 8, characterized in that, Step S4 includes the following steps: S4.1 Obtain the load information sequence and fault quantity information sequence of the target planning area in the previous historical time period, and input them into the load prediction model and fault quantity prediction model respectively to obtain the load prediction value and fault quantity prediction value at the prediction time step. The obtained load prediction value and fault quantity prediction value at the prediction time step constitute the prediction result. S4.2 Preset load thresholds and fault thresholds, compare the predicted load value with the load threshold, and compare the predicted fault number value with the fault threshold; S4.3 If the load forecast value is greater than or equal to the load threshold, or the fault number forecast value is greater than or equal to the fault threshold, the planning data for the target planning area shall be updated and adjusted; if the load forecast value is less than the load threshold and the fault number forecast value is less than the fault threshold, the planning data shall not be adjusted. S4.4 Based on the updated planning data, recalculate the data evaluation value using the data evaluation model in step S1. If the data evaluation value is not 0, return to step S2 and re-execute; if the data evaluation value is 0, stop updating the planning data and terminate the iteration.

Images