In-province power flow calculation and report generation method and platform based on digital intelligent AI

Abstract

According to the in-province power flow calculation and report generation method and platform based on the digital intelligent AI, an integrated solution of data input, model calculation and report output is formed, standardization and processizing of provincial level power flow planning work are achieved, and the planning efficiency and the planning precision are improved; the method comprises the steps of 1, predicting load electric quantity, comparing and selecting data, and forming a high-medium-low scheme; historical load data, historical electric quantity data and historical economic data are uploaded, the load electric quantity is predicted by adopting 10 methods, and 10 groups of load electric quantities are predicted and generated; calculating an average value of the 10 groups of data, taking data similar to the average value as a middle scheme, taking a group slightly higher than the middle scheme as a high scheme, and taking a group slightly lower than the middle scheme as a low scheme; step 2, power balance and power flow calculation; step 3, report generation; and calling the DeepSeek-R1-0528 model to complete the writing of the report.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power flow calculation, in particular to a provincial power flow calculation and report generation method and platform based on digital AI. BACKGROUND

[0002] At present, State Grid Corporation of China is fully promoting the construction of new power system. As the core link of the power grid planning work, it is facing unprecedented complexity, accuracy and timeliness challenges. The traditional planning mode mainly relying on manual work has been difficult to meet the requirements of the new situation. The traditional manual mode faces the challenges of large data scale, complex calculation task and long report preparation period, etc. It is difficult to efficiently process and analyze such massive, multi-source and heterogeneous data, resulting in insufficient data value mining; and the calculation task is complex and coupled. The power flow calculation is a complex optimization problem with multiple variables and multiple constraints. It needs to consider:

[0003] Variable boundary conditions: annual maximum load and power prediction are the basis, but their results need to be coupled with unit output characteristics, inter-provincial power transmission and reception protocols, load reserve rate and response scale, and new energy storage layout variables.

[0004] Strict safety constraints: the power flow limit of each 500kV partition needs to be accurately calculated and repeatedly compared and verified with the existing key section power transmission limit to ensure the safe and stable operation of the power grid.

[0005] Manual calculation is difficult to simulate and optimize thousands of scenarios, often only using simplified models and typical methods, which limits the accuracy of the planning results.

[0006] The calculation results are structured data, while the planning report is an unstructured document that needs to be interpreted and organized by humans. The conversion between the two requires a lot of manual copying, pasting and formatting work, which is not only inefficient, but also prone to human error.

[0007] Therefore, using "data + AI" technology (digital AI) to reshape the provincial power flow calculation and report generation process has become an urgent and critical research and development task. SUMMARY

[0008] The present application aims to overcome the technical defects of the prior art and provides a provincial power flow calculation and report generation method and platform based on digital AI, forming an integrated solution of "data input → model calculation → report output", realizing the standardization and process of provincial power flow planning work, and improving the planning efficiency and accuracy.

[0009] The first aspect of the present application discloses a provincial power flow calculation and report generation method based on digital AI, comprising the following steps:

[0010] Step 1, load and electricity prediction; upload historical load data, historical electricity data, and historical economic data, and use LSTM, CNN, SVM, ARIMA, elasticity coefficient method, output value unit consumption method, univariate linear regression, logarithmic regression, hyperbolic regression, and sectoral analysis method to predict the load and electricity, generating 10 groups of regional load and electricity.

[0011] Compare the data to form high, medium and low options; calculate the average of 10 sets of data, take the data that is close to the average as the medium option, take the group that is slightly higher than the medium option as the high option, and take the group that is slightly lower than the medium option as the low option.

[0012] Step 2, Power Balance and Power Flow Calculation; Input the power load of the power supply zones for the planned year, and solve for the optimal output coefficients of various units using a constrained linear equation system. First, ensure that the following identities hold in mode j:

[0013] ,

[0014] Among them, c i,j Let m be the output coefficient of the i-th type of unit in mode j. i,j Let x be the installed capacity of the i-th type of unit under mode j. j Let r be the load factor under mode j, P be the annual maximum load, and r be the load factor. j For the spinning reserve capacity under mode j, d j For inter-provincial agreement-based power transmission under the j-mode, l i,j u represents the lower limit of the output of unit i in mode j. i,j This represents the upper limit of the output of unit i in mode j;

[0015] By solving the above equation, c i,j This is then incorporated into the available capacity of various zone units under mode j to form a power balance table and a zone power surplus s without considering interactions. j,k ,

[0016] ,

[0017] Among them, c i,j,k Let m be the output coefficient of unit i in zone k under mode j. i,j,k Let P be the installed capacity of the i-th type of unit in partition k under mode j. k For the maximum annual load of partition k, r j,k The rotating reserve capacity under mode j in partition k;

[0018] The following equation stabilizes the power flow arrangement in the zones.

[0019] ,

[0020] Among them, I j,k,m To obtain the m-th power flow in k-partition j mode, the power flows are summed according to the definition of the transmission section to obtain the power flow of the section;

[0021] Step 3, report generation; call the DeepSeek-R1-0528 model to complete the report writing.

[0022] Furthermore,

[0023] The report generation in step 3 includes the following steps:

[0024] Step 3-1: Retrieve the load power results and the calculation results of power balance and power flow, and import the historical planning report into the RAG database;

[0025] Step 3-2: Perform report segmentation according to the report logic;

[0026] Step 3-3 involves combining the corresponding data for each block and the segmented architecture according to the prompt word method.

[0027] Use the DeepSeek-R1-0528 model to output;

[0028] Steps 3-4 involve integrating the report modules, combining and formatting them to complete the report writing.

[0029] Furthermore,

[0030] In step 1, the LSTM first normalizes the historical load data, historical electricity data, and historical economic data, and then constructs an input layer, several hidden layers, and an output layer. The hidden layer consists of a forget gate, an input gate, and an output gate, and its operation process uses the following formula:

[0031] Forgotten Gate:

[0032] ,

[0033] Input Gate:

[0034] ,

[0035] Update cell status:

[0036] ,

[0037] Output gates and hidden states:

[0038] ,

[0039] Where t represents the current time step, x t It is the input vector, ht It is the hidden state at the current time step, C t The cell state is represented by σ, which represents the Sigmoid activation function, and ∗ represents element-wise multiplication. W and b represent the weight matrix and bias term, respectively. The subscripts f, i, o, and C represent the parameters of the forget gate, input gate, output gate, and cell state, respectively. By continuously training and optimizing the model parameters, an optimal set of solutions is found to minimize the output error.

[0040] Furthermore,

[0041] In step 1, the CNN integrates two functions through convolution operations to form a third function, using the following formula:

[0042] ,

[0043] ,

[0044] Where s is the function formed by the convolution operation, x is the input function, w is the weight function, * is the convolution operator, I is the m×n input matrix function, K is the corresponding m×n convolution kernel function, and i and j are the row and column numbers of the elements of the convolution function s, respectively.

[0045] The CNN includes convolutional layers, pooling layers, and dense layers. The convolutional layer consists of three stages: the first stage is initiated by an activation function; the second stage rectifies the detected activation function into a linear function; and the third stage performs pooling to downsample the input data while preserving important features. The values ​​obtained from the convolutional layer are sent to the pooling layer for feature extraction. At the end of the pooling stage, the output data needs to be planarized and forwarded to the dense layer to finally obtain a one-dimensional output sequence, i.e., the prediction result.

[0046] Furthermore,

[0047] In step 1, the SVM assumes the prediction model is f(x) = wφ(x) + b, and establishes the objective function:

[0048] ,

[0049] Where C is the penalty factor, λi is the weight coefficient, ξi and ξ*i are slack variables, and N is the total number of input samples. The constraints are as follows:

[0050] ,

[0051] y i Let ε be the actual load value at time i, and let ε be the error requirement value of the function.

[0052] Furthermore,

[0053] The ARIMA in step 1, the ARIMA(p,d,q) model, is represented as follows:

[0054] ,

[0055] Where, ε t The noise is white noise; φ and θ are coefficients. The ARIMA model includes an autoregressive model part and a moving average regression part.

[0056] First, determine whether the input data is a stationary sequence, and transform non-stationary sequences into stationary ones. For each stationary sequence, construct an ARIMA(p,d,q) product model. The combination of parameters p and q is different from the standard model, and p and q are constrained to a relatively low order to avoid overfitting. Let p = 0, 1, 2, q = 0, 1. Use the AIC criterion to find the optimal model for each stationary sequence. The model with the smallest AIC value is the optimal model, therefore S0. i,t The optimal ARIMA model mathematical expression is as follows:

[0057] ,

[0058] Based on the selected optimal ARIMA model, predict the S of each preprocessed sequence. i,t The future value of , the prediction result is denoted by S i,t+τ =1, 2, ..., ΔT represents the forecast period. All ARIMA forecast results are then aggregated to obtain the final power system load forecast.

[0059] .

[0060] Furthermore,

[0061] The elasticity coefficient method in step 1 calculates the predicted electricity consumption at the end of the period by predicting a reasonable elasticity coefficient, and then inversely calculates the maximum load by using the annual maximum load utilization hours. The calculation formula is as follows:

[0062] ,

[0063] Where E0 is the predicted initial electricity consumption, E n To predict the electricity consumption at the end of the period, K is the elasticity coefficient, β is the GDP growth rate, and n is the number of years in the calculator.

[0064] The second aspect of this invention discloses a provincial power flow calculation and report generation platform based on digital AI, comprising:

[0065] The load and power prediction module is used to predict the load and power of different zones and compare data to form high, medium and low schemes;

[0066] The load and power prediction module includes a historical load data upload interface, a historical power data upload interface, a historical economic data upload interface, a load and power calculation interface, a high, medium, and low scheme interface for total social power consumption, and a high, medium, and low scheme interface for total social maximum load.

[0067] The power flow and power balance calculation module is used to calculate power flow; the power flow and power balance calculation module includes a power data upload interface, a characteristic data upload interface, an inter-provincial agreement power flow upload interface, a power flow calculation interface, a power balance table interface, and a power flow calculation diagram interface.

[0068] The report generation module is used to generate reports; the report generation module includes a template upload interface and a report generation interface.

[0069] Compared with the prior art, the beneficial effects of the present invention are:

[0070] This invention provides a method and platform for calculating and generating provincial power flow reports based on digital AI. By deeply integrating data-driven artificial intelligence technology with traditional power system planning physical models, an end-to-end intelligent planning solution is constructed, with beneficial effects reflected in improved planning efficiency and accuracy.

[0071] 1. Significantly improve the efficiency of planning work

[0072] This system has completely transformed the inefficient traditional model that relied on manual calculations and repeated adjustments to a single model. It reduces data processing and model debugging, which previously took weeks, to hours or even minutes, achieving "one-click" and "pipeline" processing in the prediction phase. It avoids the subjectivity and bias of manually setting output coefficients. The system can quickly process massive amounts of variables and complex constraints, automatically identifying the optimal solution that meets all technical conditions. It solves the core pain point of the disconnect between "calculation" and "reporting," completely freeing planners from the time-consuming document preparation work that accounts for 50%-70% of the entire project cycle. It achieves "report draft generation upon completion of calculation," reducing the report preparation cycle from weeks to hours.

[0073] 2. Significantly improve planning accuracy

[0074] Employing the approach of "combined prediction" and "multiple scheme comparison," ten algorithms capture data patterns from different dimensions (e.g., LSTM excels at time series analysis, CNN captures spatial features, and traditional models ensure stability). By averaging and adjusting for fluctuations, schemes are formed, effectively avoiding the systematic biases that may exist in a single model, making the prediction results more robust and valuable. The constraint-based linear programming model ensures the mathematical optimality of the calculation results. The model strictly considers all key constraints such as the upper and lower limits of unit output, spinning reserve, and inter-provincial agreements, making the calculated power flow limit and cross-sectional load conditions closer to the physical reality of the power grid and reducing errors caused by manual estimation.

[0075] This method seamlessly integrates four previously relatively independent processes—load forecasting, power balancing, power flow calculation, and report generation—into a complete and automated business process, forming an integrated solution of "data input → model calculation → report output." This achieves standardization and process-orientation of provincial power flow planning, and the entire integrated platform is deployed locally offline, making the method easy to replicate and promote in different regions. Attached Figure Description

[0076] Fig. 1 This is the UI interface for load and power prediction in this invention;

[0077] Fig. 2 This is the UI interface for power balance and power flow calculation in this invention;

[0078] Fig. 3 A UI interface is generated for the report of this invention. Detailed Implementation

[0079] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0080] like Figs. 1-3 As shown, it illustrates a specific embodiment of the present invention:

[0081] like Figs. 1-3As shown, the platform utilizes computer languages ​​such as Matlab and Python, and calls the large language model DeepSeek R1-0528 (for processing text content and computer language-aided programming) and AI power calculation and prediction algorithms (for power boundary prediction and electrical calculation). It has independently developed an integrated platform for power flow calculation and report generation. By inputting power boundary data and report format requirements, it sequentially realizes the functions of load power prediction, power balance and power flow calculation, and report generation. Through a human-computer interaction interface, it automatically generates power flow planning reports. The entire process of this integrated platform is deployed locally offline, which complies with the State Grid Corporation's data compliance requirements.

[0082] This platform enables routine power flow calculations. By using historical 8760 load power data, historical economic data, installed capacity boundary data, operational data, and report writing templates, it sequentially completes the functions of load power prediction, power balance and power flow calculation, and report generation, thus fulfilling routine manual tasks.

[0083] The first aspect of this invention discloses a method for calculating and generating provincial power flow based on digital AI, comprising the following steps:

[0084] Step 1, load and electricity prediction; upload historical load data, historical electricity data, and historical economic data, and use LSTM, CNN, SVM, ARIMA, elasticity coefficient method, output value unit consumption method, univariate linear regression, logarithmic regression, hyperbolic regression, and sectoral analysis method to predict the load and electricity, generating 10 groups of regional load and electricity.

[0085] Compare the data to form high, medium and low options; calculate the average of 10 sets of data, take the data that is close to the average as the medium option, take the group that is slightly higher than the medium option as the high option, and take the group that is slightly lower than the medium option as the low option.

[0086] Step 2, Power Balance and Power Flow Calculation; Input the power load of the power supply zones for the planned year, and solve for the optimal output coefficients of various units using a constrained linear equation system. First, ensure that the following identities hold in mode j:

[0087] ,

[0088] Among them, c i,j Let m be the output coefficient of the i-th type of unit in mode j. i,j Let x be the installed capacity of the i-th type of unit under mode j. j Let r be the load factor under mode j, P be the annual maximum load, and r be the load factor. j For the spinning reserve capacity under mode j, d j For inter-provincial agreement-based power transmission under the j-mode, l i,ju represents the lower limit of the output of unit i in mode j. i,j This represents the upper limit of the output of unit i in mode j;

[0089] By solving the above equation, c i,j This is then incorporated into the available capacity of various zone units under mode j to form a power balance table and a zone power surplus s without considering interactions. j,k ,

[0090] ,

[0091] Among them, c i,j,k Let m be the output coefficient of unit i in zone k under mode j. i,j,k Let P be the installed capacity of the i-th type of unit in partition k under mode j. k For the maximum annual load of partition k, r j,k The rotating reserve capacity under mode j in partition k;

[0092] The following equation stabilizes the power flow arrangement in the zones.

[0093] ,

[0094] Among them, I j,k,m To obtain the m-th power flow in k-partition j mode, the power flows are summed according to the definition of the transmission section to obtain the power flow of the section;

[0095] Step 3, report generation; call the DeepSeek-R1-0528 model to complete the report writing.

[0096] Furthermore, the report generation in step 3 includes the following steps:

[0097] Step 3-1: Retrieve the load power results and the calculation results of power balance and power flow, and import the historical planning report into the RAG database;

[0098] Step 3-2: Perform report segmentation according to the report logic;

[0099] Step 3-3 involves combining the corresponding data for each block and the segmented architecture according to the prompt word method.

[0100] Use the DeepSeek-R1-0528 model to output;

[0101] Steps 3-4 involve integrating the report modules, combining and formatting them to complete the report writing.

[0102] The above process is programmed using Python and Matlab, and compiled sequentially into sub-functions zhuanhuan, fenge, zhuanxie, and zhengheg to implement the above functions. The above algorithm is then sent to the main UI interface via callback, with the report template as a Word document as input. The report is written by calling the above algorithm, and the completed report is then opened via a connection.

[0103] Furthermore, in step 1, the LSTM first normalizes the historical load data, historical electricity data, and historical economic data, and constructs an input layer, several hidden layers, and an output layer. The hidden layer consists of a forget gate, an input gate, and an output gate, and its operation process uses the following formula:

[0104] Forgotten Gate:

[0105] ,

[0106] Input Gate:

[0107] ,

[0108] Update cell status:

[0109] ,

[0110] Output gates and hidden states:

[0111] ,

[0112] Where t represents the current time step, x t It is the input vector, h t It is the hidden state at the current time step, C t The cell state is represented by σ, which represents the Sigmoid activation function, and ∗ represents element-wise multiplication. W and b represent the weight matrix and bias term, respectively. The subscripts f, i, o, and C represent the parameters of the forget gate, input gate, output gate, and cell state, respectively. By continuously training and optimizing the model parameters, an optimal set of solutions is found to minimize the output error.

[0113] The LSTM (Long Short-Term Memory) network first normalizes the historical 8760 load data. Through continuous training and optimization of the model parameters, it finds an optimal solution to minimize the output error.

[0114] Furthermore, in step 1, the CNN integrates two functions through convolution operations to form a third function, using the following formula:

[0115] ,

[0116] ,

[0117] Where s is the function formed by the convolution operation, x is the input function, w is the weight function, * is the convolution operator, I is the m×n input matrix function, K is the corresponding m×n convolution kernel function, and i and j are the row and column numbers of the elements of the convolution function s, respectively.

[0118] The CNN includes convolutional layers, pooling layers, and dense layers. The convolutional layer consists of three stages: the first stage is initiated by an activation function; the second stage rectifies the detected activation function into a linear function; and the third stage performs pooling to downsample the input data while preserving important features. The values ​​obtained from the convolutional layer are sent to the pooling layer for feature extraction. At the end of the pooling stage, the output data needs to be planarized and forwarded to the dense layer to finally obtain a one-dimensional output sequence, i.e., the prediction result.

[0119] Furthermore, in step 1, the SVM assumes the prediction model is f(x) = wφ(x) + b, and establishes the objective function:

[0120] ,

[0121] Where C is the penalty factor, λi is the weight coefficient, ξi and ξ*i are slack variables, and N is the total number of input samples. The constraints are as follows:

[0122] ,

[0123] y i Let ε be the actual load value at time i, and let ε be the error requirement value of the function.

[0124] Furthermore, the ARIMA(p,d,q) model in step 1 is represented as follows:

[0125] ,

[0126] Where, ε t The noise is white noise; φ and θ are coefficients. The ARIMA model includes an autoregressive model part and a moving average regression part.

[0127] First, determine whether the input data is a stationary sequence, and transform non-stationary sequences into stationary ones. For each stationary sequence, construct an ARIMA(p,d,q) product model. The combination of parameters p and q is different from the standard model, and p and q are constrained to a relatively low order to avoid overfitting. Let p = 0, 1, 2, q = 0, 1. Use the AIC criterion to find the optimal model for each stationary sequence. The model with the smallest AIC value is the optimal model, therefore S0. i,t The optimal ARIMA model mathematical expression is as follows:

[0128] ,

[0129] Based on the selected optimal ARIMA model, predict the S of each preprocessed sequence. i,t The future value of , the prediction result is denoted by S i,t+τ =1, 2, ..., ΔT represents the forecast period. All ARIMA forecast results are then aggregated to obtain the final power system load forecast.

[0130] .

[0131] Furthermore, the elasticity coefficient method in step 1 calculates the predicted electricity consumption at the end of the period by predicting a reasonable elasticity coefficient, and then inversely calculates the maximum load by using the annual maximum load utilization hours. The calculation formula is as follows:

[0132] ,

[0133] Where E0 is the predicted initial electricity consumption, E n To predict the electricity consumption at the end of the period, K is the elasticity coefficient, β is the GDP growth rate, and n is the number of years in the calculator.

[0134] Output-based consumption method: This method predicts the total annual electricity consumption of a region by forecasting the electricity consumption per unit of local GDP output, and then uses the annual maximum load utilization hours to infer the maximum load.

[0135] Univariate linear regression: Univariate linear regression is a linear prediction method, typically used in scenarios where the trend of change is relatively stable. It can be equivalent to the following formula:

[0136] ,

[0137] Where x is the independent variable and y is the dependent variable. The values ​​of a and b are obtained through historical data. Here, the independent variable is the year, and the dependent variable is the maximum electricity consumption (load) of the entire society.

[0138] ,

[0139] ,

[0140] in,

[0141] ,

[0142] t is the sequence number of the time series, x t For the t-th input time series, y t This is the t-th output time series.

[0143] Logarithmic regression and hyperbolic regression: Univariate nonlinear regression analysis is mainly applied to scenarios where the predicted value changes at different levels over different periods. By selecting a regression function (model) suitable for the trend of predicted value changes, as shown in Table 1, the predicted value for a future period can be predicted. By transforming the independent and dependent variables, it can be equivalent to univariate linear regression analysis.

[0144] Table 1 Regression Analysis Model

[0145]

[0146] Where x is the independent variable, y is the dependent variable, a and b are the relevant parameters, x' is the transformed independent variable, and y' is the equivalent dependent variable.

[0147] Sectoral analysis method: The composition of historical electricity consumption is divided into sectors (industries) for forecasting, and the forecast results of each sector are added together to obtain the electricity consumption value for the forecast year. The maximum load is then inferred from the annual maximum load utilization hours.

[0148] The above 10 algorithms are used to capture data patterns from different dimensions (e.g., LSTM excels at temporal data, CNN captures spatial features, and traditional models ensure stability). By taking the average and adjusting for fluctuations, a scheme is formed, effectively avoiding the systematic bias that may exist in a single model, making the prediction results more robust and valuable. The algorithms are compiled into sub-functions such as sub_normal, sub_ARIMA, sub_SVM, sub_LSTM, and sub_CNN through Matlab programming, and high, medium, and low schemes are formed through the comparison sub-function sub_compare. The above algorithms are called into the main UI interface via callbacks. The 8760 load data, electricity data, and economic data from the past 10 years are used as input in Excel format. The above algorithms are called to perform calculations. The data that is close to the average value output by the above methods is taken as the medium scheme, the slightly higher one is taken as the high scheme, and the slightly lower one is taken as the low scheme.

[0149] This invention, through the empowerment of "AI+", drastically reduces the time required for calculation and report writing from 30 days for a single person to just 1 hour, achieving a significant leap in efficiency. It powerfully validates the feasible path of "AI+ technology efficiently and systematically replacing mechanized and procedural human work," providing strong momentum for the continuous liberation and development of productivity in power planning and even broader fields.

[0150] Compared to the traditional "table-based" method, AI prediction is more accurate: AI prediction and calculation are trained on long-term historical hourly data waveforms of the region and optimized for solution. The output results have an error of less than 2% compared to historical actual values ​​and government planning boundaries, meeting the actual needs of planning work. Compared to the traditional "table-based" method, AI calculation is more efficient: Manual calculation is usually based on existing massive amounts of planning boundary data, requiring data sorting, statistical tabulation, and then manually finding the optimal solution and drawing diagrams, a tedious and time-consuming process. AI calculation, with its optimization algorithms and automatic drawing technology, can reduce the time required for the above process from 15 days to seconds. Compared to traditional manual writing, AI writing is more scientific: AI writing includes all the logical framework and content of the original report, while adding data characteristic analysis and in-depth mining through data fusion and verification. Compared to traditional manual writing, AI writing is more intelligent: It uses the reasoning capabilities of large models to analyze data, evaluate and suggest the current situation, and assist in decision-making.

[0151] This invention also discloses a provincial power flow calculation and report generation platform based on digital AI, including:

[0152] The load and power prediction module is used to predict the load and power of different zones and compare data to form high, medium and low schemes;

[0153] The load and power prediction module includes a historical load data upload interface, a historical power data upload interface, a historical economic data upload interface, a load and power calculation interface, a high, medium, and low scheme interface for total social power consumption, and a high, medium, and low scheme interface for total social maximum load.

[0154] The power flow and power balance calculation module is used to calculate power flow; the power flow and power balance calculation module includes a power data upload interface, a characteristic data upload interface, an inter-provincial agreement power flow upload interface, a power flow calculation interface, a power balance table interface, and a power flow calculation diagram interface.

[0155] The report generation module is used to generate reports; the report generation module includes a template upload interface and a report generation interface.

[0156] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention. These changes involve related technologies well known to those skilled in the art, and all of them fall within the protection scope of the present invention.

[0157] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. A method for in-province power flow calculation and report generation based on digital AI, characterized in that, Comprise the following steps: Step 1, load power prediction; upload historical load data, historical power data, historical economic data, adopt LSTM, CNN, SVM, ARIMA, elasticity coefficient method, output value unit consumption method, linear regression, logarithmic regression, hyperbolic regression, department analysis method respectively to predict load power, and generate 10 groups of partition load power; Compare and select data to form high, medium and low schemes; obtain the average value of the 10 groups of data, and take the data close to the average value as the medium scheme, a group slightly higher than the medium scheme as the high scheme, and a group slightly lower than the medium scheme as the low scheme; Step 2, power balance and power flow calculation; input the power supply partition load power of the planning year, and solve the optimal output coefficient of each type of unit through the method of constraint-based linear equation set, so that the following identity holds in j mode: ， wherein c i,j is the output coefficient of the i-th type of unit in the j-th mode, m i,j is the installed capacity of the i-th type of unit in the j-th mode, x j is the load coefficient in the j-th mode, P is the annual maximum load, r j is the spinning reserve capacity in the j-th mode, d j is the inter-provincial agreement export in the j-th mode, l i,j is the output lower limit of the i-th type of unit in the j-th mode, u i,j is the output upper limit of the i-th type of unit in the j-th mode. By solving the c i,j And into the j way under each type of partition unit available capacity to form a power balance table and do not consider the interaction of the partition power surplus s j,k , ， wherein c i,j,k is the output coefficient of the i-th type of unit in the k-th partition in the j-th mode, m i,j,k is the installed capacity of the i-th type of unit in the k-th partition in the j-th mode, P k is the annual maximum load of the k-th partition, r j,k is the spinning reserve capacity in the j-th mode in the k-th partition; Make the partition power flow arrangement stable through the following equation, ， where I j,k,m is the mth power flow in the kth partition j scheme, and the power flow of the section is obtained by adding the power flows according to the definition of the transmission section. Step 3, report generation; call DeepSeek-R1-0528 model to complete the writing of the report.

2. The digit intelligence AI based inter-state power flow calculation and report generation method as claimed in claim 1 wherein, The report generation in the step 3 comprises the following steps: Step 3-1, call the load power result and the calculation result of power balance and power flow, and import the historical planning report into the RAG database; Step 3-2, perform report logic segmentation; Step 3-3, according to the prompt word method, combine each corresponding data and the segmented architecture to Call DeepSeek-R1-0528 model for output; Step 3-4, integrate the report module, combine layout, format conversion, and complete the writing of the report.

3. The digit intelligence AI based inter-state power flow calculation and report generation method as claimed in claim 2 wherein, The LSTM in the step 1 firstly normalizes the historical load data, historical power data and historical economic data, constructs an input layer, a plurality of hidden layers and an output layer, the hidden layers are composed of a forgetting gate, an input gate and an output gate, and the operation process adopts the following formula: Forgetting gate: ， Input gate: ， Update unit state: ， Output gate and hidden state: ， where t represents the current time step, x t is the input vector, h t is the hidden state of the current time step, C t is the cell state, σ denotes the Sigmoid activation function, * denotes element-wise multiplication, W and b denote the weight matrix and bias term, respectively, and the subscripts f, i, o, C represent the parameters of the forget gate, input gate, output gate, and cell state, respectively. Through continuous training and optimization of the model parameters, an optimal solution is found to minimize the output error.

4. The digitAI-based interregional power flow calculation and report generation method of claim 1, wherein, The CNN in the step 1 integrates two functions to form a third function through convolution operation, and adopts the following formula: ， ， Wherein, s is a function formed by convolution operation, x is an input function, w is a weight function, * is a convolution operator, I is an input matrix function of m*n, K is a convolution kernel function corresponding to m*n, i and j are respectively the row number and column number of the element of the convolution function s; The CNN comprises a convolution layer, a pooling layer and a dense layer, wherein the convolution layer is composed of three stages, the first stage is started through an activation function, the second stage rectifies the detected activation function to a linear function, and the third stage performs a pooling operation to reduce the sampling of input data while retaining important features of the data, the values obtained from the convolution layer are sent to the pooling layer for feature extraction, and at the end of the pooling stage, the output data needs to be flattened and forwarded to the dense layer, and finally a one-dimensional output sequence, i.e. a prediction result, is obtained.

5. The digitAI-based interregional power flow calculation and report generation method of claim 1, wherein, The SVM in the step 1 assumes that the prediction model is f(x)=wφ(x)+b, and establishes an objective function: ， Wherein, C is a penalty factor, λi is a weight coefficient, ξi and ξ*i are relaxation variables, N is the total number of input samples, and the constraint condition is: ， y i is the true value of the load at the i-th time instant, and ε is the error requirement value of the function.

6. The digitAI-based interregional power flow calculation and report generation method of claim 1, wherein, The ARIMA, ARIMA (p, d, q) model in the step 1 is expressed as: ， wherein ε t is white noise; φ, θ are coefficients, and the ARIMA model comprises an autoregressive model part and a moving average regression part; First, the input data is judged whether it is a stationary sequence, and the non-stationary sequence is converted into a stationary sequence. For each stationary sequence, an ARIMA(p, d, q) product model is constructed. The combination of parameters p and q is different, and p and q are limited to a relatively low order to avoid overfitting. Let p = 0, 1, 2, q = 0, 1, and find the optimal model of each stationary sequence using the AIC criterion. The model with the smallest AIC value is the optimal model, so the optimal ARIMA model of S i,t The mathematical expression of the optimal ARIMA model is as follows: ， According to the corresponding optimal ARIMA model selected, the future values of each pre-processed sequence S i,t are predicted, and the prediction results are represented as S i,t+τ =1, 2, …, ΔT; where ΔT is the prediction period, and all the ARIMA prediction results are summarized to obtain the final power system load prediction result: 。 7. The digitAI-based interregional power flow calculation and report generation method of claim 1, wherein, The elasticity coefficient method in the step 1 calculates the final predicted power by predicting a reasonable elasticity coefficient, and calculates the maximum load by using the annual maximum load utilization hours, and the calculation formula is as follows: ， Where E0 is the initial electricity consumption, E is the final electricity consumption, K is the elasticity coefficient, β is the GDP growth rate, and n is the number of years. n Where E0 is the initial electricity consumption, E is the final electricity consumption, K is the elasticity coefficient, β is the GDP growth rate, and n is the number of years.

8. A digital AI-based inter-provincial power flow calculation and report generation platform, characterized in that, The method for calculating and generating the intra-provincial power flow and report based on digital AI according to any one of claims 1-8 comprises: A load power prediction module is configured to predict the partition load power, and form high, medium and low schemes by comparing and selecting data; The load power prediction module comprises a historical load data uploading interface, a historical power data uploading interface, a historical economic data uploading interface, a load power calculation interface, a high, medium and low scheme interface of the total social power consumption, and a high, medium and low scheme interface of the total social maximum load; A power flow and power balance calculation module is configured to calculate the power flow, and the power flow and power balance calculation module comprises a power source data uploading interface, a characteristic data uploading interface, a provincial interconnection agreement power flow uploading interface, a power flow calculation interface, a power balance table interface and a power flow calculation diagram interface; A report generation module is configured to generate a report, and the report generation module comprises a template uploading interface and a report generation interface.

Images