Natural gas peak regulation method and system based on spatio-temporal information

Abstract

The invention belongs to the technical field of natural gas peak regulation, particularly relates to a natural gas peak regulation method and system based on spatio-temporal information, and aims to solve the problem that a large-range and more comprehensive natural gas peak regulation mode is lacked in the prior art. The method comprises the following steps: acquiring a second index thermodynamic diagram and inputting the second index thermodynamic diagram into a first neural network model to obtain the natural gas consumption of a target area in a first statistical period as a first demand quantity; before the first statistical period starts, natural gas of the first demand quantity is conveyed to a central peak regulation system of the target area; obtaining a fourth index thermodynamic diagram, inputting the fourth index thermodynamic diagram into the second neural network model, and predicting the natural gas consumption of each sub-region of the target region in the first statistical period as a second demand; and before the first statistical period starts, the natural gas of each second demand quantity is conveyed to the corresponding secondary peak regulation system. According to the method, large-range and more comprehensive natural gas peak regulation is realized through a grading direct prediction mode.

Description

Technical Field

[0001] The present invention belongs to the technical field of natural gas peak shaving, and in particular relates to a natural gas peak shaving method and system based on time-space information. Background Art

[0002] In recent years, with the proposal of carbon peak and carbon neutrality goals, the importance of natural gas as a clean energy source has been increasing day by day. Since natural gas use has peak periods, the amount of natural gas used during peak periods far exceeds the maximum amount of natural gas that the pipeline can bear during peak periods. Therefore, it is necessary to deliver sufficient natural gas to storage facilities in the target area in advance during non-peak periods to adjust the supply and demand of natural gas during peak periods. This method of using the redundant delivery volume of non-peak pipelines to deliver the natural gas required during peak periods in advance is called natural gas peak shaving.

[0003] The natural gas pipeline peak-shaving method provided by the existing system or method uses the SCADA system (Supervisory Control And Data Acquisition). Although the system realizes functions such as data acquisition, facility control, and measurement parameter adjustment, in the face of changes in demand for natural gas in regions and time domains across the country, as well as the contradiction of unbalanced natural gas supply, human factors still need to intervene in local areas, and there is a lack of a large-scale and more comprehensive peak-shaving method. The peak-shaving method based on human intervention has low operating efficiency based on experience, and the peak-shaving result is difficult to achieve the optimal peak-shaving solution for the current situation. After peak-shaving, there will be a mismatch between supply and demand. Therefore, it is difficult to meet the natural gas peak-shaving demand in large areas, and it is difficult to achieve a low-cost, high-efficiency natural gas peak-shaving effect. Based on this, the present invention provides a natural gas peak-shaving method based on spatiotemporal information. Summary of the invention

[0004] In order to solve the above problems in the prior art, that is, to solve the problem that the prior art lacks a large-scale and more comprehensive natural gas peak-shaving method, and the peak-shaving result is difficult to achieve the optimal peak-shaving solution for the current situation, and there is a problem of mismatch between supply and demand after peak-shaving, the present invention proposes a natural gas peak-shaving method based on spatiotemporal information, which includes:

[0005] The second index heat map is obtained and input into the trained first neural network model to predict the natural gas consumption of the target area in the first statistical period as the first demand; the second index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of the target area in the first statistical period on the geographical information of the target area; the target area includes multiple sub-areas;

[0006] Before the start of the first statistical period, the first demanded amount of natural gas is delivered to the central peak-shaving system of the target area; the central peak-shaving system is connected to a plurality of secondary peak-shaving systems, each sub-area has a secondary peak-shaving system, and each secondary peak-shaving system includes a plurality of pipeline nodes and a plurality of peak-shaving equipment;

[0007] A fourth index heat map is obtained and input into the trained second neural network model to predict the natural gas consumption of each sub-region of the target area within the first statistical period as the second demand; the fourth index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of each sub-region in the target area within the first statistical period on the geographic information;

[0008] Before the first statistical period starts, the natural gas of each second demand amount is respectively delivered to the corresponding secondary peak-shaving system.

[0009] In some preferred embodiments, the training methods of the first neural network model and the second neural network model are:

[0010] The training method of the first neural network model is:

[0011] Acquire data information of multiple statistical periods within a first set time period of the target area; the data information includes geographic information, meteorological information, electricity consumption information, natural gas consumption information and economic information; the first set time period includes multiple statistical periods;

[0012] Superimposing the meteorological information, electricity consumption information and economic information of the target area in each statistical period onto the geographic information to obtain a plurality of first index heat map samples;

[0013] Overlaying the natural gas consumption information of the target area in each statistical period onto the geographic information to obtain multiple natural gas consumption heat maps;

[0014] Taking the plurality of first index thermogram samples as input, and taking the natural gas usage thermogram corresponding to each of the first index thermogram samples as label data, training the pre-constructed first neural network model until a trained first neural network model is obtained;

[0015] The training method of the second neural network model is:

[0016] A third index heat map obtained by superimposing meteorological information, electricity consumption information and economic information of multiple sub-areas of the target area within the first set time period onto geographic information is used as input, and a natural gas consumption heat map of the target area within the first set time period is used as label data for training to obtain a trained second neural network model.

[0017] In some preferred embodiments, after the second required amount of natural gas is respectively delivered to the corresponding secondary peak shaving system, the method further includes:

[0018] At the end of the first statistical period, the actual natural gas consumption of all sub-areas of the target area is obtained as the first usage amount; the length of the second set time period is equal to the length of one statistical period;

[0019] Acquire meteorological information, electricity consumption information and economic information in the target area during the first statistical period; the meteorological information, electricity consumption information and economic information in the target area include meteorological information, electricity consumption information and economic information of a sub-area of ​​each target area;

[0020] superimposing the meteorological information, electricity consumption information and economic information of each sub-region in the target region during the first statistical period onto the geographic information of the target region to obtain a plurality of fifth index heat maps;

[0021] The second neural network model is retrained with the multiple fifth index heat maps as input, the multiple first usage amounts as label data, and the natural gas demand amount as output.

[0022] In some preferred embodiments, after the second required amount of natural gas is respectively delivered to the corresponding secondary peak shaving system, the method further includes:

[0023] Acquire the actual natural gas consumption of the target area within a second set time period as the second usage; the second set time period is the same as the starting time point of the first statistical period; the length of the second set time period is equal to the length of multiple statistical periods;

[0024] Acquiring meteorological information, electricity consumption information, and economic information in the target area during the second set time period;

[0025] superimposing the meteorological information, electricity consumption information and economic information of the target area in the second set time period onto the geographic information of the target area to obtain a sixth index heat map;

[0026] The first neural network model is retrained with the sixth index heat map as input, the second usage as label data, and the natural gas demand as output.

[0027] In some preferred embodiments, after the second required amount of natural gas is respectively delivered to the corresponding secondary peak shaving system, the method further includes:

[0028] Obtaining target flow values ​​of all pipe network nodes in each secondary peak shaving system within the first statistical period; the sum of the target flow values ​​of all pipe network nodes in the secondary peak shaving system is equal to the second demand corresponding to the secondary peak shaving system;

[0029] Input multiple target flow values ​​into the trained dual-network deep learning algorithm model, perform data search operations on the on / off status of all peak-shaving devices in each secondary peak-shaving system, obtain the first state of all peak-shaving devices in each secondary peak-shaving system, and the second predicted flow value of all pipe network nodes in each secondary peak-shaving system in the first state;

[0030] At the beginning of the first statistical period, the plurality of peak-shaving devices are controlled to switch to the corresponding first state, and natural gas of each second predicted flow value is respectively delivered to the corresponding pipeline network node.

[0031] In some preferred embodiments, the dual network deep learning algorithm model is trained by:

[0032] Obtaining the on / off status of all peak-shaving devices in each secondary peak-shaving system within a third set time period as a second state; the third set time period is a historical time period and includes multiple statistical periods; the second state includes the on / off status of all peak-shaving devices in the secondary peak-shaving system within a statistical period;

[0033] Obtaining the flow values ​​of all the pipe network nodes in each secondary peak-shaving system within the third set time period to obtain a plurality of first flow values; the first flow values ​​include the flow values ​​of all the pipe network nodes in each secondary peak-shaving system within a statistical period;

[0034] Using a plurality of the second states as training samples, inputting the pre-built dual-network deep learning algorithm model, and obtaining a first predicted flow value;

[0035] Based on the first predicted flow value and its corresponding label data, the loss value of the dual network deep learning algorithm model is calculated, and the model parameters are updated; the label data is the first flow value of all pipe network nodes in each secondary peak-shaving system;

[0036] The dual-network deep learning algorithm model is trained in a loop until the loss value is less than a first set value, and then the training is stopped to obtain a trained dual-network deep learning algorithm model;

[0037] The loss function of the dual network deep learning algorithm model during training is:

[0038]

[0039] Among them, L represents the loss function, q 1pi represents the first predicted flow value of the i-th pipe network node, q 1i represents the first flow value of the i-th pipe network node, and n represents the number of pipe network nodes.

[0040] In some preferred embodiments, the dual-network deep learning algorithm model includes a filling network and a training network;

[0041] The filling network has a first input dimension and a first output dimension; the number of dimensions of the first input dimension is numerically equal to the number of effective peak-shaving devices; the number of dimensions of the first output dimension is numerically equal to the sum of the number of all peak-shaving devices in all secondary peak-shaving systems; the effective peak-shaving devices are peak-shaving devices that work normally within a statistical period; the first input dimension is used to input the on / off status of all effective peak-shaving devices;

[0042] The training network has a second input dimension and a second output dimension; the number of dimensions of the second input dimension is numerically equal to the sum of the number of all peak-shaving devices in all secondary peak-shaving systems; the number of dimensions of the second output dimension is numerically equal to the number of all pipeline nodes in the secondary peak-shaving system; the second input dimension corresponds to the first output dimension one by one and is interconnected; the result output by the second output dimension is the predicted flow value of natural gas at each pipeline node.

[0043] In some preferred embodiments, the target flow value is predicted by:

[0044] A10, randomly initializing the value table to obtain the Nth value table, including: randomly setting the flow value of each pipe network node in the secondary peak-shaving system as the Nth target flow value; randomly setting the adjustment amount of each pipe network node as the Nth adjustment amount; the initial value of N is set to 1;

[0045] A20, calculating the Nth objective function value of each secondary peak load regulation system pipeline network node according to the Nth target flow value;

[0046] The objective function is:

[0047] Among them, F represents the objective function, Q represents the Nth target flow value, D represents the second demand, c represents the gas adjustment fee, cs represents the gas flow ratio in adjacent unit time, k represents the number of adjustments, and a1, a2, a3, and a4 represent the weights of each item respectively;

[0048] A30, obtaining an N+1th target flow value of each pipe network node according to the Nth adjustment amount and the Nth target flow value;

[0049] A40, calculating the N+1th objective function value of each secondary peak load regulation system pipeline network node according to the N+1th target flow value;

[0050] A50, when the N+1th objective function value is less than the Nth objective function value, a reward value is assigned to the adjustment value corresponding to the Nth adjustment amount, and the process jumps to A60; otherwise, the process jumps directly to A60; the initial reward value of each adjustment value is 0, and the reward values ​​assigned to the same adjustment value multiple times can be accumulated;

[0051] A60, updating the value table, including: selecting the N+1th adjustment amount of each pipe network node according to the set rule to obtain the N+1th value table;

[0052] A70, determine whether the Nth objective function value satisfies the first set condition, if not, set N=N+1 and jump to A20, otherwise use the current Nth objective function value as the target flow value.

[0053] In some preferred embodiments, the gas conditioning fee is calculated as follows:

[0054]

[0055] Among them, A represents the number of peak load-shaving equipment, B represents the number of gas extraction times, and H j represents the amount of natural gas regulated by the jth peak load shaving device, R j It represents the cost of peak-shaving energy per unit gas volume when the j-th peak-shaving device performs peak-shaving, U g Indicates the gas production cost per unit time, T g Indicates the gas sampling time.

[0056] In a second aspect of the present invention, a natural gas peak shaving system based on time-space information is proposed, the system comprising:

[0057] The first demand prediction module is configured to obtain a second index heat map and input the trained first neural network model to predict the natural gas consumption of the target area in the first statistical period as the first demand; the second index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of the target area in the first statistical period on the geographic information of the target area; the target area includes multiple sub-areas;

[0058] A delivery module is configured to deliver the first demanded amount of natural gas to the central peak-shaving system of the target area before the start of the first statistical period; the central peak-shaving system is connected to a plurality of secondary peak-shaving systems, each sub-area has a secondary peak-shaving system, and each secondary peak-shaving system includes a plurality of pipeline network nodes and a plurality of peak-shaving devices;

[0059] The second demand prediction module is configured to obtain a fourth index heat map and input the trained second neural network model to predict the natural gas consumption of each sub-region of the target area within the first statistical period as the second demand; the fourth index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of each sub-region in the target area within the first statistical period on the geographic information;

[0060] The peak-shaving control module is configured to deliver each second demand amount of natural gas to the corresponding secondary peak-shaving system before the start of the first statistical period.

[0061] Beneficial effects of the present invention:

[0062] The present invention obtains meteorological information, electricity consumption information and economic information of the target area in the first statistical period, and superimposes them on the geographical information of the target area to obtain a second index heat map. The second index heat map is input into the first neural network model, and the first neural network model is used to predict the natural gas consumption of the target area in the first statistical period to obtain the first demand. The first demand of natural gas is input into the central peak-shaving system before the first statistical period. Then, a third index heat map is made, and the second neural network model is input, and the natural gas demand of each sub-area is predicted by the second neural network model to obtain multiple second demands. The second demand of natural gas is respectively delivered to the corresponding secondary peak-shaving system before the first statistical period. The predicted data of the first statistical period of the target area is used in combination with the neural network model to predict the natural gas demand of the target area. Since the neural network is trained by historical data, the meteorological information, electricity consumption information and economic information in the same statistical period are input to predict the natural gas demand; the method of predicting the natural gas demand by using the neural network is closer to the optimal solution of peak-shaving than human intervention, which is convenient for automatic control of peak-shaving and improves operation efficiency. The hierarchical prediction and redistribution method can better adapt to the overall natural gas peak-shaving demand in a large range, while meeting the demand for precise natural gas peak-shaving in a small range. The above-mentioned hierarchical direct prediction method can also reduce the computational complexity, reduce the calculation time of the neural network, and be more suitable for short-term prediction, thereby realizing a low-cost and high-efficiency peak-shaving method. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings.

[0064] Figure 1 It is a flow chart of a natural gas peak shaving method based on time-space information according to an embodiment of the present invention;

[0065] Figure 2It is a structural schematic diagram of a central peak-shaving system and a secondary peak-shaving system in one embodiment of the present invention;

[0066] Figure 3 A schematic diagram of the structure of a dual-network deep learning algorithm model according to an embodiment of the present invention;

[0067] Figure 4 A schematic diagram of cascade feedback regulation of a natural gas peak-shaving system based on spatiotemporal information according to an embodiment of the present invention;

[0068] Among them: 1. Filling network; 2. Training network; 3. First input dimension; 4. First output dimension; 5. Second input dimension; 6. Second output dimension; 7. Central peak-shaving facilities; 8. Secondary peak-shaving facilities; 9. Pipeline network nodes; 10. Peak-shaving equipment; 11. Gas terminals. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It should be noted that, in the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0071] A natural gas peak shaving method based on time-space information in a first embodiment of the present invention is as follows: Figure 1 As shown, the method includes:

[0072] The second index heat map is obtained and input into the trained first neural network model to predict the natural gas consumption of the target area in the first statistical period as the first demand; the second index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of the target area in the first statistical period on the geographical information of the target area; the target area includes multiple sub-areas;

[0073] Before the start of the first statistical period, the first demanded amount of natural gas is delivered to the central peak-shaving system of the target area; the central peak-shaving system is connected to a plurality of secondary peak-shaving systems, each sub-area has a secondary peak-shaving system, and each secondary peak-shaving system includes a plurality of pipeline nodes and a plurality of peak-shaving equipment;

[0074] A fourth index heat map is obtained and input into the trained second neural network model to predict the natural gas consumption of each sub-region of the target area within the first statistical period as the second demand; the fourth index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of each sub-region in the target area within the first statistical period on the geographic information;

[0075] Before the first statistical period starts, the natural gas of each second demand amount is respectively delivered to the corresponding secondary peak-shaving system.

[0076] In order to more clearly illustrate the natural gas peak-shaving method based on time-space information of the present invention, each step in an embodiment of the method of the present invention is described in detail below with reference to the accompanying drawings.

[0077] Using neural networks to predict natural gas demand is closer to the optimal solution for peak load regulation than human intervention, which facilitates automatic control of peak load regulation and improves operating efficiency. The method of graded prediction and redistribution can better adapt to the overall natural gas peak load demand in a large range, while meeting the demand for precise natural gas peak load regulation in a small range. The above-mentioned graded direct prediction method can also reduce the computational complexity, reduce the calculation time of the neural network, and be more suitable for short-term prediction, thereby realizing a low-cost and high-efficiency peak load regulation method.

[0078] In the following embodiments, the training process of the first neural network model and the second neural network model is first described in detail, and then the process of natural gas peak shaving by a natural gas peak shaving method based on spatiotemporal information is described in detail.

[0079] 1. Training process of the first neural network model and the second neural network model

[0080] In the present invention, the model structures of the first neural network model and the second neural network model are the same or different. In this embodiment, the first neural network model and the second neural network model are preferably constructed based on a convolutional neural network; in other embodiments, they can be selected according to actual conditions, for example, they can be constructed by a deep learning neural network.

[0081] For the first neural network model, the training process is as follows:

[0082] Acquire data information of multiple statistical periods within a first set time period of the target area; the data information includes geographic information, meteorological information, electricity consumption information, natural gas consumption information and economic information; the first set time period includes multiple statistical periods;

[0083] In this embodiment, if the target area is set to a country, then the sub-area is set to a province. The statistical period is preferably set to one year, and the first set time period is preferably set to the length of time from many years ago to the present. If the target area is set to a province, then the sub-area is set to a city. The statistical period is preferably set to one month, and the first set time period is preferably set to one year, two years or multiple months. If the target area is set to a city, then the sub-area is set to the districts and counties included in the city. In other embodiments, the setting of the statistical period is not limited to natural dates. For example, the period from mid-June to mid-August can be used as a statistical period, that is, a sliding window method is used to predict the natural gas demand on a quarterly basis.

[0084] Geographic information includes the natural area and administrative divisions of the target area; meteorological information includes cloud data, temperature data and wind speed in the target area during the first set time period. Changes in meteorological information will lead to changes in the climate in the target area. For example, lower temperatures will cause local residents to have heating needs, which will indirectly affect the demand for natural gas, so climate change will also affect the use of natural gas;

[0085] The electricity consumption information includes residential electricity consumption and industrial electricity consumption in the target area within the first set time period. The local population of the target area will be directly reflected in the residential electricity consumption, and the industrial development level will be directly reflected in the industrial electricity consumption. Residents' daily dining and industrial production will directly or indirectly consume natural gas, so the total population and industrial development level will also directly affect the demand for natural gas.

[0086] The economic information includes the GDP value and the average natural gas unit price of the target area during the first set time period. The economic development of the target area and the natural gas unit price will affect the willingness of residents to use natural gas, and thus directly affect the demand for natural gas. For example, if the economic situation in a certain area is poor and the natural gas unit price is high, the willingness of local residents to use natural gas will be relatively low; if the economic situation in another area is good and the natural gas unit price is low, the willingness of local residents to use natural gas will be relatively high.

[0087] In summary, considering meteorological information, electricity consumption information and economic information can consider the factors affecting natural gas demand from multiple angles, and thus more accurately predict the natural gas demand in the target area in the future; finally, the peak-shaving method described in the previous embodiment is used to more reasonably allocate the natural gas demand in the target area.

[0088] Superimposing the meteorological information, electricity consumption information and economic information of the target area in each statistical period onto the geographic information to obtain a plurality of first index heat map samples;

[0089] In this embodiment, the geographic information is automatically interpreted to obtain the feature information (including rivers, forests, agricultural areas, industrial areas, and residential areas, etc.) contained in the geographic information of the target area; based on the feature information and the meteorological information, electricity information, and economic information of the target area within the first set time period, the first index heat map sample is superimposed. The production process of the second index heat map and the third index heat map described below is the same. The index heat map contains the geographic information of the target area, and after superposition, each area in the geographic information is marked with the meteorological information, electricity information, and economic information of the area. After the heat map is input into the neural network model, the information marked by the corresponding area can be directly read.

[0090] Overlaying the natural gas consumption information of the target area in each statistical period onto the geographic information to obtain multiple natural gas consumption heat maps;

[0091] The plurality of first index thermogram samples are used as input, and the natural gas usage thermogram corresponding to each of the first index thermogram samples is used as label data to train the pre-constructed first neural network model until a trained first neural network model is obtained.

[0092] The second neural network model, its training method is:

[0093] The third index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of multiple sub-regions of the target area in the first set time period onto the geographic information as input, and the natural gas consumption heat map of the target area in the first set time period (i.e., superimposing the natural gas consumption information of each sub-region onto the geographic information corresponding to each sub-region) is used as label data for training to obtain a trained second neural network model. The first neural network model and the second neural network model can use the same neural network (e.g., convolutional neural network, deep neural network, etc.), and the process of the second neural network model is similar to the training process of the first neural network model, which will not be described in detail here.

[0094] During the training process of the first neural network model and the second neural network model, the model will continuously output the predicted value of natural gas demand in the target area. As the training progresses, the predicted value becomes closer to the actual value in the label data.

[0095] 2. A natural gas peak-shaving method based on spatiotemporal information

[0096] In the present invention, when forecasting the natural gas demand for the whole country in the next year, the country is taken as the largest unit, and the lower units are provinces or cities. The central peak-shaving system forecasts the natural gas demand of each province or city, and the sum of the total is the forecast demand for natural gas in the whole country. When the total gas supply is determined, the system algorithm is used to optimize the allocation of resources to each province or city.

[0097] After the secondary peak-shaving system receives the allocated natural gas, it can treat the overall peak-shaving facilities in the province as the central peak-shaving system and the peak-shaving facilities in each city as the secondary peak-shaving system for redistribution. In the process of redistribution, the demand of all gas terminals 11 in each district and county can be further distributed by treating the peak-shaving facilities in the city as the central peak-shaving system and the peak-shaving facilities in each district and county as the secondary peak-shaving system.

[0098] The above-mentioned multi-level peak-shaving method can not only be applied to macro-peak-shaving at the national level, but also to districts and counties, and even to the distribution of natural gas between multiple communities. In terms of the peak-shaving cycle, the national peak-shaving cycle can be set to one year; the statistical cycle for districts and counties can be set to one month; the statistical cycle for communities can be set to one week or one day. Combining the above contents with each other, it is also possible to first predict the natural gas demand at the national level for one year; when it comes to districts and counties, a natural gas demand forecast for the next month is made for each district and county every month within a year. This type of peak-shaving method, which is both hierarchical and periodic, can take into account both macro-peak-shaving at the national level and precise peak-shaving for specific districts and counties.

[0099] The central peak-shaving system and the secondary peak-shaving system use two neural network models for prediction, which can predict the data in a hierarchical manner, reduce the amount of calculation, and improve the prediction efficiency of the neural network model. The specific peak-shaving process is as follows:

[0100] The second index heat map is obtained and input into the trained first neural network model to predict the natural gas consumption of the target area in the first statistical period as the first demand; the second index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of the target area in the first statistical period on the geographical information of the target area; the target area includes multiple sub-areas;

[0101] In this embodiment, an index heat map of a statistical period to be predicted (i.e., the first statistical period) is obtained and recorded as a second index heat map. The second index heat map is input into the trained first neural network model to predict the natural gas consumption of the target area in the first statistical period as the first demand.

[0102] Before the start of the first statistical period, the first demanded amount of natural gas is delivered to the central peak-shaving system of the target area; the central peak-shaving system is connected to a plurality of secondary peak-shaving systems, each sub-area has a secondary peak-shaving system, and each secondary peak-shaving system includes a plurality of pipeline nodes and a plurality of peak-shaving devices; Figure 2 shown.

[0103] Peak-shaving equipment includes: compressors in valve rooms of long-distance pipeline stations, underground gas storage peak-shaving equipment, gas tank peak-shaving equipment, liquefied petroleum gas peak-shaving equipment, liquefied natural gas peak-shaving equipment, and all other equipment used to control the flow size of pipeline network nodes;

[0104] The pipeline network nodes are the intersections of pipelines. There are a large number of natural gas transmission pipelines, and the pipeline connections are intricate. Each secondary peak-shaving system has a large number of pipeline network nodes;

[0105] Obtaining a fourth index heat map and inputting it into a trained second neural network model to predict the natural gas consumption of each sub-area of ​​the target area within the first statistical period as a second demand;

[0106] Before the first statistical period begins, the natural gas of each second demand is delivered to the corresponding secondary peak-shaving system. Figure 2 As shown, the present invention uses the first neural network model to predict the natural gas demand in the first statistical period of the target area, and then delivers the first demand of natural gas to the central peak-shaving facility 7 (also referred to as the central peak-shaving system); and then uses the second neural network model to predict the natural gas demand of each secondary peak-shaving facility 8 (also referred to as the secondary peak-shaving system) to obtain the second demand. Then, by controlling the opening and closing states of multiple peak-shaving devices 10 of the secondary peak-shaving system 8 and adjusting the flow values ​​of multiple pipeline nodes 9, the second demand of natural gas is distributed to each gas terminal 11.

[0107] After the natural gas of each second demand amount is respectively delivered to the corresponding secondary peak-shaving system, the method further includes:

[0108] At the end of the first statistical period, the actual natural gas consumption of all sub-areas of the target area is obtained as the first usage amount; the length of the second set time period is equal to the length of one statistical period;

[0109] Acquire meteorological information, electricity consumption information and economic information in the target area during the first statistical period; the meteorological information, electricity consumption information and economic information in the target area include meteorological information, electricity consumption information and economic information of a sub-area of ​​each target area;

[0110] superimposing the meteorological information, electricity consumption information and economic information of each sub-region in the target region during the first statistical period onto the geographic information of the target region to obtain a plurality of fifth index heat maps;

[0111] The second neural network model is retrained with the multiple fifth index heat maps as input, the multiple first usage amounts as label data, and the natural gas demand amount as output.

[0112] After completing the natural gas transmission of a statistical cycle, the total natural gas consumption in the target area in the statistical cycle that just ended is counted, and then the meteorological information, electricity information and economic information of the target area in this statistical cycle are counted; the total natural gas consumption, meteorological information, electricity information and economic information in this statistical cycle are used as historical data to make an index heat map, and then input into the second neural network model as training data to retrain the second neural network model. The above method enables the neural network model to automatically update and adjust according to the actual natural gas consumption status of the target area, and can consider the impact of accidental over-consumption or under-consumption on the actual prediction results, so that the predicted value is closer to the true value.

[0113] Since the change rate of natural gas demand in a large range is small, it is possible to consider retraining only the second neural network model, which can reduce the training cost of the model and save the time of building the model.

[0114] After the natural gas of each second demand amount is respectively delivered to the corresponding secondary peak-shaving system, the method further includes:

[0115] Acquire the actual natural gas consumption of the target area within a second set time period as the second usage; the second set time period is the same as the starting time point of the first statistical period; the length of the second set time period is equal to the length of multiple statistical periods;

[0116] Acquiring meteorological information, electricity consumption information, and economic information in the target area during the second set time period;

[0117] superimposing the meteorological information, electricity consumption information and economic information of the target area in the second set time period onto the geographic information of the target area to obtain a sixth index heat map;

[0118] The first neural network model is retrained with the sixth index heat map as input, the second usage as label data, and the natural gas demand as output; and the neural network model obtained by the retraining is used as the first neural network model again.

[0119] In the present invention, the first neural network model is retrained at a longer interval, and the second neural network model is retrained at a shorter interval. For example, when making predictions for the whole country, the first neural network model is retrained once a year; when making predictions for each province, the second neural network model is retrained once a month. Since the rate of change of natural gas demand in a small area is large, and the rate of change of natural gas demand in a large area is small; adaptively retraining the first neural network model at a longer interval and retraining the second neural network model at a shorter interval can make the prediction results of the first neural network model maintain the stability of the national natural gas supply, while making the second neural network model adapt to the natural gas demand with large short-term changes in a small area; the combination of the two can make the prediction of natural gas demand more intelligent and adapt to a wider range of uses.

[0120] In the process of retraining the first neural network model and the second neural network model, it is defaulted that the geographic information remains unchanged.

[0121] In addition, during the training process, the meteorological information, electricity consumption information, and economic information of the target area and all its sub-areas within the first set time period are respectively superimposed on the geographic information of the target area to produce a sixth index heat map.

[0122] The neural network model is trained using the sixth index heat map to obtain a third neural network model. In this embodiment, the third neural network model is preferably constructed based on a convolutional neural network model. In other embodiments, it can also be constructed based on other neural network models.

[0123] In the process of predicting the national natural gas demand, due to the large total amount, the central peak-shaving system will not actually store the national natural gas consumption for one year. Therefore, the first neural network model and the second neural network model use the same neural network method to directly predict multiple sub-areas of the target area separately, obtain multiple second demands, and then sum up the multiple second demands to obtain the first demand.

[0124] This method can skip the steps of predicting the national natural gas demand and storing natural gas in the central peak-shaving system when predicting the national natural gas demand, and at the same time know the total national natural gas demand. Then, in the actual transportation process, each second demand of natural gas is directly transported to the corresponding province. When facing the peak-shaving of a large target area, it can simplify the peak-shaving operation, reduce the amount of calculation, and improve the peak-shaving efficiency.

[0125] The present invention uses the meteorological information, electricity consumption information and economic information of the first statistical period to predict the natural gas demand of the target area in the first statistical period. Since the neural network is trained by historical data, the meteorological information, electricity consumption information and economic information in the same statistical period are input to predict the natural gas demand; the method of using the neural network model to predict the natural gas demand is closer to the optimal solution for peak load regulation than the method of human intervention, which facilitates automatic control of peak load regulation and improves operation efficiency.

[0126] The prediction of the next cycle of meteorological information, electricity consumption information and economic information of the target area all belong to the existing technology. The technical solution of this application can directly use the existing technology to obtain the above information during the implementation process. For example, the patent document with publication number CN115130788A details an electricity consumption prediction system based on big data analysis; the patent document with publication number CN112749792A details a wind speed prediction method based on the BP algorithm; the patent document with publication number CN115270453A details a cloud movement trajectory prediction method and system; the patent document with publication number CN115510767A details a regional temperature prediction method based on a deep spatiotemporal network; the patent document with publication number CN103440536A details a regional macroeconomic forecasting model method;

[0127] In fact, the method of using the meteorological information, electricity consumption information and economic information of the first statistical period to predict the natural gas demand in the target area in the first statistical period aims to decompose part of the calculation amount into other algorithms, reduce the calculation complexity of the first neural network model and the second neural network model, enable the first neural network model and the second neural network model to make rapid predictions, and make the model more suitable for the prediction of natural gas demand in a short statistical period.

[0128] Specifically, when it is necessary to predict the natural gas demand of a specific district or county in the next week, sometimes it is only necessary to directly query the local temperature, wind speed, cloud layer, GDP value, residential electricity consumption, industrial electricity consumption and natural gas unit price. Then, a thermal index map is made and input into the second neural network model, which can completely eliminate the time consumption of predicting the local temperature, wind speed, cloud layer, GDP value, residential electricity consumption, industrial electricity consumption and natural gas unit price, and quickly predict the natural gas demand of the district or county in the next week. When there is a shortage of natural gas in the district or county, the local natural gas demand can be quickly predicted in a short period of time, saving time for the allocation of natural gas.

[0129] In other embodiments, the target area's recent meteorological information, electricity consumption information, and economic information can also be used to predict the natural gas demand of the target area in the first statistical period; this can reduce the error between the predicted natural gas demand and the actual natural gas usage. Since the natural gas usage change rate between different statistical periods in a large area is small, this type of prediction method with a smaller error is more suitable for prediction in a large area to ensure the stability of large-scale natural gas peak regulation.

[0130] The secondary peak-shaving system contains multiple pipe network nodes and multiple peak-shaving facilities. Due to the complex structure of the pipe network, the flow rate of each pipe network node is related to each peak-shaving device in the secondary peak-shaving system, that is, the opening and closing of a peak-shaving device will affect the flow rate at all nodes of the entire pipe network. Therefore, a dual-network deep learning algorithm model is required to search the opening and closing status of all peak-shaving devices in the secondary peak-shaving system, and then find the opening and closing status of all peak-shaving devices when each pipe network node is at the target flow value.

[0131] Therefore, after the second required amount of natural gas is respectively delivered to the corresponding secondary peak-shaving system, the method further includes:

[0132] Obtaining target flow values ​​of all pipe network nodes in each secondary peak shaving system within the first statistical period; the sum of the target flow values ​​of all pipe network nodes in the secondary peak shaving system is equal to the second demand corresponding to the secondary peak shaving system;

[0133] In this embodiment, the value table is updated through the idea of ​​reinforcement learning to obtain the target flow value. Specifically: initialize the value table, that is, randomly set the initial flow value and regulation amount of each network node; calculate the initial value of the secondary peak-shaving system objective function; perform an iteration, during which the flow value of each network node changes according to the regulation amount; calculate the value of the objective function again; compare the numerical values ​​of the two objective functions before and after, and when the value of the objective function decreases, give a reward; otherwise, no reward or a penalty; update the value table and perform the next iteration; until the objective function no longer decreases after multiple consecutive iterations; wherein the regulation amount is to increase or decrease the flow value of the network node by a certain amount, whether the flow value increases or decreases, and the specific value of the regulation is obtained according to the set rules. ; The adjustment amounts of different pipe network nodes in the same iteration cycle are different, and the increase and decrease situations are also different. The adjustment amounts of the adjustment amounts in different iteration cycles may also be different; the size of the reward value given is related to the value of the objective function reduction and the number of iterations. The more it is reduced, the more rewards are given; as the number of iterations increases, the reward value corresponding to a one-unit reduction in the objective function value will also increase. The process of the value table includes: updating the flow value table, using the Nth target flow value and the Nth adjustment amount to add to get the N+1th target flow value; updating the reward value table, when the N+1th objective function value is less than the Nth objective function value, assign a reward value to the adjustment value corresponding to the Nth adjustment amount; otherwise, no reward is given. Specifically:

[0134] A10, randomly initializing the value table to obtain the Nth value table, including: randomly setting the flow value of each pipe network node in the secondary peak-shaving system as the Nth target flow value; randomly setting the adjustment amount of each pipe network node as the Nth adjustment amount; the initial value of N is set to 1;

[0135] In the present invention, the value range of the Nth target flow value is [0, μ]; the specific value range of the Nth adjustment amount is [-μ, μ], and μ represents the maximum flow rate that can be transported by the pipeline.

[0136] A20, calculating the Nth objective function value of each secondary peak load regulation system pipeline network node according to the Nth target flow value;

[0137] The objective function is:

[0138]

[0139] Among them, F represents the objective function, Q represents the Nth target flow value, D represents the second demand, c represents the gas adjustment fee, cs represents the gas flow ratio in adjacent unit time, k represents the number of adjustments, a1, a2, a3, and a4 represent the weights of each item respectively; the sum of a1, a2, a3, and a4 is equal to 100%. The gas flow ratio cs in adjacent unit time is directly detected by the detection device, and when cs=1, it means that the gas flow in adjacent unit time has not changed. The number of adjustments k is specifically the number of adjustments required for the state combination of all current peak-shaving devices to the state combination of all peak-shaving devices corresponding to the Nth target flow value. The gas adjustment fee is calculated as follows:

[0140]

[0141] Among them, A represents the number of peak load-shaving equipment, B represents the number of gas extraction times, and H j represents the amount of natural gas regulated by the jth peak load shaving device, R j represents the energy cost per unit gas volume when the j-th peak-shaving device performs peak-shaving (for example, when a gas storage is supplying gas, energy is consumed when the gas from the gas storage is transported to the pipeline, which represents the energy cost of the gas storage for every cubic meter of gas supplied), U g Indicates the gas production cost per unit time, T g Indicates the gas sampling time;

[0142] A30, obtaining the N+1th target flow value of each pipe network node according to the Nth adjustment amount and the Nth target flow value; that is, the Nth adjustment amount is added to the Nth target flow value to obtain the N+1th target flow value;

[0143] A40, calculating the N+1th objective function value of each secondary peak load regulation system pipeline network node according to the N+1th target flow value;

[0144] A50, when the N+1th objective function value is less than the Nth objective function value, a reward value is assigned to the adjustment value corresponding to the Nth adjustment amount, and the process jumps to A60; otherwise, the process jumps directly to A60; the initial reward value of each adjustment value is 0, and the reward values ​​assigned to the same adjustment value multiple times can be accumulated;

[0145] A60, updating the value table, including: selecting the N+1th adjustment amount of each pipe network node according to a set rule to obtain the N+1th value table; the set rule is specifically:

[0146] A first numerical range is set; a first set threshold is selected within the first numerical range; a numerical value is randomly selected within the first numerical range as a first random number; when the first random number is greater than the first set threshold, the value of the N+1th adjustment amount is randomly selected; when the first random number is less than the first set threshold, the adjustment amount with the largest reward value in the historical adjustment amounts is selected as the N+1th adjustment amount; in the present invention, the setting rule is preferably an ε-greed rule; wherein the first numerical range is set to [0,1]; and the first set threshold ε is set to 0.9.

[0147] A70, determine whether the Nth objective function value satisfies the first setting condition, if not, set N=N+1, jump to A20, otherwise use the current Nth objective function value as the target flow value; the first setting condition is that the objective function continues to decrease for multiple cycles. Generally, it is preferably set to no longer decrease for ten cycles;

[0148] The above method enables the algorithm to find the optimal solution nearby more quickly. After multiple rounds of training, the algorithm can find the optimal solution nearby with as few steps as possible, and then find the target flow value more quickly. The larger the first set threshold, the greater the probability of selecting the maximum reward value, and the algorithm will quickly approach the optimal solution nearby, so the algorithm will quickly approach the optimal solution nearby, and it is easy to fall into the local optimal solution. The smaller the first set threshold, the greater the probability of random exploration of the algorithm, the longer the exploration time, and the greater the probability of jumping out of the local optimal solution compared with the former. Therefore, it is necessary to set the value of the first set threshold according to actual needs. For example, when facing the national natural gas demand forecast, due to the long cycle, a smaller first set threshold is selected, and the algorithm has sufficient time to explore and then seek the global optimal solution; it can ensure the stability of the national natural gas supply. When predicting areas with temporary shortages of natural gas, a larger first set threshold is selected, which can quickly find the local optimal solution and save time for the allocation of natural gas.

[0149] In addition, the value table in the present invention includes a flow value table and a reward value table:

[0150] Table 1 Nth target flow value Pipeline network node 1 Pipeline network node 2 Pipeline network node 3 ... The first target flow value 10 10 10 ... The second target flow value 11 12 9 ... The third target flow value 12 11 8 ... ... ... ... ...

[0151] Table 2

[0152] Table 1 is a flow value table, and Table 2 is a reward value table. In Table 2, the reward value column is initially 0. As the iteration proceeds, the reward value corresponding to the adjustment value assigned to the reward value is accumulated, indicating that during the iteration process, selecting this adjustment amount can reduce the objective function.

[0153] The above method enables the algorithm to find the optimal solution nearby more quickly. After multiple rounds of training, the algorithm can find the optimal solution nearby with as few steps as possible, and then find the target flow value more quickly.

[0154] Input multiple target flow values ​​into the trained dual-network deep learning algorithm model, perform data search operations on the on / off status of all peak-shaving devices in each secondary peak-shaving system, obtain the first state of all peak-shaving devices in each secondary peak-shaving system, and the second predicted flow value of all pipe network nodes in each secondary peak-shaving system in the first state;

[0155] This is because in the actual transportation process, the actual natural gas supply may be insufficient, and there may also be some surplus natural gas stored in the secondary peak-shaving system. Therefore, in the actual process of supplying natural gas, it is difficult to supply strictly according to demand.

[0156] Specifically, when the predicted demand is less than the actual available supply and there is no surplus, the gas will be delivered directly according to the predicted demand. When the predicted demand is less than the actual available supply and there is a surplus, the total amount of natural gas actually delivered shall be equal to the predicted demand minus the surplus.

[0157] When the actual supply available is less than the predicted demand, the present embodiment is implemented: a dual-network deep learning algorithm model is used to perform data search operations on the on and off states of all peak-shaving equipment in the secondary peak-shaving system, and a state combination of the on and off of multiple peak-shaving facilities is obtained, so that the flow value of each pipeline node under this state combination is closest to the target flow value as a whole.

[0158] Under the actual supply and demand relationship, when the supply is insufficient, there will be no margin in each secondary peak-shaving system; therefore, this application assumes that when the supply is insufficient, there will be no margin in the secondary peak-shaving system. Under the above peak-shaving method, when the supply is insufficient, it is ensured that all gas terminals 11 have gas available, and there will be no situation where some gas terminals 11 have no gas available.

[0159] In the present invention, the dual-network deep learning algorithm model includes a filling network and a training network; the filling network and the training network are preferably constructed based on the ResNet network; in other embodiments, other network models can also be selected according to actual conditions to construct the filling network and the training network, such as a deep learning network.

[0160] The filling network 1 has a first input dimension 3 and a first output dimension 4; the number of dimensions of the first input dimension 3 is numerically equal to the number of effective peak-shaving devices; the number of dimensions of the first output dimension 4 is numerically equal to the sum of the number of all peak-shaving devices in all secondary peak-shaving systems; the effective peak-shaving devices are peak-shaving devices that work normally within a statistical period; the first input dimension 3 is used to input the on / off status of all effective peak-shaving devices;

[0161] The training network 2 has a second input dimension 5 and a second output dimension 6; the dimension number of the second input dimension 5 is numerically equal to the sum of the number of all peak-shaving devices in all secondary peak-shaving systems; the dimension number of the second output dimension 6 is numerically equal to the number of all pipeline nodes in the secondary peak-shaving system; the second input dimension 5 corresponds to the first output dimension 4 and is interconnected; the result output by the second output dimension 6 is the predicted flow value of natural gas at each pipeline node.

[0162] The structure of the dual network deep learning algorithm model is as follows Figure 3 As shown, the number of dimensions of the first input dimension 3, the first output dimension 4 and the second input dimension 5 are all equal to the number of peak-shaving devices 10 in the secondary peak-shaving system 8, and the number of dimensions of the second output dimension 6 is equal to the number of all pipeline nodes 9 in the secondary peak-shaving facility 8 (also referred to as the secondary peak-shaving system).

[0163] Since there may be faulty peak-shaving devices 10 in the secondary peak-shaving system 8 within a statistical period, the fault time and duration of the fault are difficult to know through investigation, and the flow value of each pipeline node 9 is related to the opening and closing status of each peak-shaving device 10, it is difficult to eliminate the impact of the faulty peak-shaving devices 10 on the flow value of the pipeline node 9.

[0164] However, after investigation, it is possible to know which specific peak-shaving devices 10 have failed, that is, the number of failed peak-shaving devices 10. Therefore, when inputting the on / off status of the peak-shaving devices 10, only the status of the peak-shaving devices 10 that work normally is input, and the combination of the on / off status of the peak-shaving devices 10 is fitted using the filling network.

[0165] Specifically: the flow value of the network node 9 corresponding to the state output by the first output dimension 4 is used as the flow value corresponding to the input state; the flow value of the network node 9 corresponding to the state output by the first output dimension 4 is used as the flow value corresponding to the output state. After filling the network 1, the flow value corresponding to the input state is equal to the flow value corresponding to the output state, and the output state is equivalent to the output state. In this way, the influence of the faulty peak-shaving equipment 10 on the flow value of the network node 9 is eliminated. Finally, the equivalent state of the peak-shaving equipment 10 is used as input and input into the training network for training.

[0166] Taking 100 peak-shaving devices and 200 pipe network nodes as an example, assuming that 2 peak-shaving devices fail within a statistical period, when inputting the first input dimension, only the on / off status of 98 normal peak-shaving devices is input, and after the on / off status of the equivalent 100 peak-shaving devices is output in the first output dimension, the on / off status of the equivalent 100 peak-shaving devices is input into the training network for training, and the first predicted flow value of 200 pipe network nodes is output. This can effectively eliminate the influence of the faulty peak-shaving device 10 on the flow value of the pipe network node 9, and ensure that the predicted second predicted flow value is closer to the true value.

[0167] The dual network deep learning algorithm model, its training method is:

[0168] Obtaining the on / off status of all peak-shaving devices in each secondary peak-shaving system within a third set time period as a second state; the third set time period is a historical time period and includes multiple statistical periods; the second state includes the on / off status of all peak-shaving devices in the secondary peak-shaving system within a statistical period;

[0169] Obtaining the flow values ​​of all the pipe network nodes in each secondary peak-shaving system within the third set time period to obtain a plurality of first flow values; the first flow values ​​include the flow values ​​of all the pipe network nodes in each secondary peak-shaving system within a statistical period;

[0170] Using multiple second states as training samples, the pre-built dual-network deep learning algorithm model is input to obtain a first predicted flow value; the first flow value is the actual flow value of all pipeline nodes in the secondary peak-shaving system within a third set time period; the first predicted flow is the predicted flow value of all pipeline nodes in the secondary peak-shaving system within the third set time period.

[0171] Based on the first predicted flow value and its corresponding label data, the loss value of the dual network deep learning algorithm model is calculated, and the model parameters are updated; the label data is the first flow value of all pipe network nodes in each secondary peak-shaving system;

[0172] The dual network deep learning algorithm model is trained cyclically until the loss value is less than the first set value, then the training is stopped to obtain a trained dual network deep learning algorithm model; the first set value can be set to any value between 2% and 5% according to the actual situation of the target area. According to historical data, when the flow value of the pipe network node changes greatly between different statistical periods in the target area, the first set value can be set to 5%; when the flow value of the pipe network node changes little between different statistical periods in the target area, the first set value can be set to 2%.

[0173] The dual network deep learning algorithm model outputs the first predicted flow value in the prediction stage. The state combination is shown in Table 1. When the dual network deep learning algorithm model is input for training, the actual input value is a matrix with one row and multiple columns, for example, [1,0,1,...].

[0174] The loss function of the dual network deep learning algorithm model during training is:

[0175]

[0176] Among them, L represents the loss function, q 1pi represents the first predicted flow value of the i-th pipe network node, q 1i represents the first flow value of the i-th pipe network node, and n represents the number of pipe network nodes.

[0177] Input multiple target flow values ​​into the trained dual-network deep learning algorithm model, and perform data search operations on the on / off status of all peak-shaving devices in each secondary peak-shaving system. The specific steps of the data search operation include:

[0178] S10, obtaining target flow values ​​of all pipe network nodes in the secondary peak-shaving system;

[0179] S20, randomly selecting a state combination of all unmarked peak-shaving devices in the secondary peak-shaving system, and inputting the state combination into the dual-network deep learning algorithm model to obtain a fifth flow value;

[0180] S30, calculating the error between the fifth flow value and the target flow value to obtain a plurality of error values;

[0181] S40, when there is an error value greater than or equal to N% (preferably 5% in the present invention), proceed to S50; otherwise proceed to S60;

[0182] S50, marking the state combination of all current peak-shaving devices in the secondary peak-shaving system, and returning to S20;

[0183] S60, taking the fifth flow value as the predicted flow value (when all error values ​​are less than 5%); combining the states of all peak-shaving devices in the current secondary peak-shaving system as the first state; and stopping the data search operation.

[0184] Among them, a state combination includes the Boolean value of the on / off state of all peak-shaving devices in the secondary peak-shaving system; that is, 1 represents on and 0 represents off. The details are shown in Table 3 (on / off state table of peak-shaving devices):

[0185] Table 3 The first peak load device The second peak load device The third peak load equipment ... 1 0 1 ...

[0186] The Boolean value in Table 3 only represents one of the situations of opening and closing the peak load device. When the natural gas supply is too small and the data search operation cannot find a state that satisfies the error condition of less than 5%, the following steps are performed:

[0187] Calculate the error value; calculate the variance of the error value; after the data search is completed, select the fifth flow value corresponding to a state combination with the smallest variance as the predicted flow value, and use the state of all peak-shaving devices in the current secondary peak-shaving system as the first state;

[0188] The above peak load regulation method can comprehensively consider and reasonably allocate natural gas when the natural gas supply is in short supply, and ensure that all gas terminals 11 have gas available in a short period of time, and then wait for the emergency allocation of natural gas.

[0189] In the actual natural gas transaction process, the gas supply contract between the two parties usually includes the following gas supply plans:

[0190] When the amount of natural gas used per unit time is lower than the minimum value agreed in the gas supply contract, billing shall be based on the minimum value agreed in the gas supply contract, and the excess amount exceeding the minimum value agreed in the gas supply contract shall be billed separately.

[0191] Under such a plan, natural gas demanders cannot obtain natural gas at the maximum pipeline delivery volume in all time periods due to cost considerations, nor can they save costs by shutting down most of the peak-shaving equipment to reduce the amount of natural gas obtained. That is, no matter how much natural gas the demander uses per unit time, its cost will not be less than the cost of using natural gas at the minimum value agreed in the gas supply contract. Therefore, natural gas demanders will face the problem of high costs or low natural gas utilization.

[0192] To solve the above problem, assume that m represents the number of peak-shaving devices turned on, and n represents the total number of peak-shaving devices. When selecting the state combination of peak-shaving devices, make the ratio of m to n greater than the first ratio to balance low cost and high utilization of natural gas. That is, under the premise that the flow value of each pipeline node during the statistical period is not less than the minimum value agreed in the gas supply contract, minimize additional natural gas consumption, and at the same time meet the natural gas demand of each gas terminal to improve natural gas utilization.

[0193] The specific value of the first ratio may be adjusted according to the minimum value actually agreed in the contract; for example, in some cases the first ratio is set to 95%.

[0194] Finally, at the beginning of the first statistical period, the plurality of peak-shaving devices are controlled to switch to the corresponding first state, and natural gas of each second predicted flow value is respectively delivered to the corresponding pipeline network node.

[0195] In addition, the present invention provides a schematic diagram of cascade feedback regulation of a natural gas peak load regulation system based on time and space information, such as Figure 4 As shown. The spatiotemporal information data management system is used to obtain data information of the statistical period, and the data processing system, the central peak-shaving system and the secondary peak-shaving system are used for demand prediction and delivery. The spatiotemporal information data management system obtains data and inputs the data into the data processing system to predict the demand for natural gas. The natural gas demand is input into the central peak-shaving system to predict the gas supply of each pipeline node, and then the predicted gas supply of each pipeline node is input into the secondary peak-shaving system. The secondary peak-shaving performs the peak-shaving means of the above embodiment according to the predicted opening and closing status of the peak-shaving facilities. After the statistical period ends, the actual consumption of natural gas at the pipeline nodes is monitored and fed back to the spatiotemporal information data management system, and the neural network model in the central peak-shaving system is retrained so that it is constantly updated with the actual natural gas consumption.

[0196] A natural gas peak shaving system based on time-space information according to a second embodiment of the present invention comprises:

[0197] The first demand prediction module 100 is configured to obtain a second index heat map and input the trained first neural network model to predict the natural gas consumption of the target area in the first statistical period as the first demand; the second index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of the target area in the first statistical period on the geographical information of the target area; the target area includes multiple sub-areas;

[0198] The delivery module 200 is configured to deliver the first demanded amount of natural gas to the central peak-shaving system of the target area before the start of the first statistical period; the central peak-shaving system is connected to a plurality of secondary peak-shaving systems, each sub-area has a secondary peak-shaving system, and each secondary peak-shaving system includes a plurality of pipeline network nodes and a plurality of peak-shaving devices;

[0199] The second demand prediction module 300 is configured to obtain a fourth index heat map and input the trained second neural network model to predict the natural gas consumption of each sub-region of the target area within the first statistical period as the second demand; the fourth index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of each sub-region in the target area within the first statistical period on the geographic information;

[0200] The peak shaving control module 400 is configured to deliver the second required amounts of natural gas to the corresponding secondary peak shaving system before the start of the first statistical period.

[0201] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0202] It should be noted that the natural gas peak-shaving system based on spatiotemporal information provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.

[0203] An electronic device according to a third embodiment of the present invention comprises at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned natural gas peak-shaving method based on spatiotemporal information.

[0204] A fourth embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned natural gas peak-shaving method based on time-space information.

[0205] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the electronic device and computer-readable storage medium described above can refer to the corresponding process in the aforementioned method example and will not be repeated here.

[0206] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0207] The terms "first", "second", "third", etc. are used to distinguish similar objects rather than to describe or indicate a particular order or sequence.

[0208] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A natural gas peak load regulation method based on spatiotemporal information, characterized in that: The method includes: The second index heat map is obtained and input into the trained first neural network model to predict the natural gas consumption of the target area in the first statistical period as the first demand; the second index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of the target area in the first statistical period on the geographical information of the target area; the target area includes multiple sub-areas; Before the start of the first statistical period, the first demanded amount of natural gas is delivered to the central peak-shaving system of the target area; the central peak-shaving system is connected to a plurality of secondary peak-shaving systems, each sub-area has a secondary peak-shaving system, and each secondary peak-shaving system includes a plurality of pipeline nodes and a plurality of peak-shaving equipment; A fourth index heat map is obtained and input into the trained second neural network model to predict the natural gas consumption of each sub-region of the target area within the first statistical period as the second demand; the fourth index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of each sub-region in the target area within the first statistical period on the geographic information; Before the first statistical period starts, the natural gas of each second demand amount is respectively delivered to the corresponding secondary peak-shaving system.

2. A natural gas peak load shaving method based on spatiotemporal information according to claim 1, characterized in that: The training methods of the first neural network model and the second neural network model are: The training method of the first neural network model is: Acquire data information of multiple statistical periods within a first set time period of the target area; the data information includes geographic information, meteorological information, electricity consumption information, natural gas consumption information and economic information; the first set time period includes multiple statistical periods; Superimposing the meteorological information, electricity consumption information and economic information of the target area in each statistical period onto the geographic information to obtain a plurality of first index heat map samples; Overlaying the natural gas consumption information of the target area in each statistical period onto the geographic information to obtain multiple natural gas consumption heat maps; Taking the plurality of first index thermogram samples as input, and taking the natural gas usage thermogram corresponding to each of the first index thermogram samples as label data, training the pre-constructed first neural network model until a trained first neural network model is obtained; The training method of the second neural network model is: A third index heat map obtained by superimposing meteorological information, electricity consumption information and economic information of multiple sub-areas of the target area within the first set time period onto geographic information is used as input, and a natural gas consumption heat map of the target area within the first set time period is used as label data for training to obtain a trained second neural network model.

3. The natural gas peak load shaving method based on time-space information according to claim 1, characterized in that: After the second required amount of natural gas is respectively delivered to the corresponding secondary peak-shaving system, the method further includes: At the end of the first statistical period, the actual natural gas consumption of all sub-areas of the target area is obtained as the first usage amount; the length of the second set time period is equal to the length of one statistical period; Acquire meteorological information, electricity consumption information and economic information in the target area during the first statistical period; the meteorological information, electricity consumption information and economic information in the target area include meteorological information, electricity consumption information and economic information of a sub-area of ​​each target area; superimposing the meteorological information, electricity consumption information and economic information of each sub-region in the target region during the first statistical period onto the geographic information of the target region to obtain a plurality of fifth index heat maps; The second neural network model is retrained with the multiple fifth index heat maps as input, the multiple first usage amounts as label data, and the natural gas demand amount as output.

4. The natural gas peak load shaving method based on time-space information according to claim 3 is characterized in that: After the natural gas of each second demand amount is respectively delivered to the corresponding secondary peak-shaving system, the method further includes: Acquire the actual natural gas consumption of the target area within a second set time period as the second usage; the second set time period is the same as the starting time point of the first statistical period; the length of the second set time period is equal to the length of multiple statistical periods; Acquiring meteorological information, electricity consumption information, and economic information in the target area during the second set time period; superimposing the meteorological information, electricity consumption information and economic information of the target area in the second set time period onto the geographic information of the target area to obtain a sixth index heat map; The first neural network model is retrained with the sixth index heat map as input, the second usage as label data, and the natural gas demand as output.

5. A natural gas peak load shaving method based on time-space information according to claim 4, characterized in that: After the natural gas of each second demand amount is respectively delivered to the corresponding secondary peak-shaving system, the method further includes: Obtaining target flow values ​​of all pipe network nodes in each secondary peak shaving system within the first statistical period; the sum of the target flow values ​​of all pipe network nodes in the secondary peak shaving system is equal to the second demand corresponding to the secondary peak shaving system; Input multiple target flow values ​​into the trained dual-network deep learning algorithm model, perform data search operations on the on / off status of all peak-shaving devices in each secondary peak-shaving system, obtain the first state of all peak-shaving devices in each secondary peak-shaving system, and the second predicted flow value of all pipe network nodes in each secondary peak-shaving system in the first state; At the beginning of the first statistical period, the plurality of peak-shaving devices are controlled to switch to the corresponding first state, and natural gas of each second predicted flow value is respectively delivered to the corresponding pipeline network node.

6. A natural gas peak load shaving method based on time-space information according to claim 5, characterized in that: The dual network deep learning algorithm model, its training method is: Obtaining the on / off status of all peak-shaving devices in each secondary peak-shaving system within a third set time period as a second state; the third set time period is a historical time period and includes multiple statistical periods; the second state includes the on / off status of all peak-shaving devices in the secondary peak-shaving system within a statistical period; Obtaining the flow values ​​of all the pipe network nodes in each secondary peak-shaving system within the third set time period to obtain a plurality of first flow values; the first flow values ​​include the flow values ​​of all the pipe network nodes in each secondary peak-shaving system within a statistical period; Using a plurality of the second states as training samples, inputting the pre-built dual-network deep learning algorithm model, and obtaining a first predicted flow value; Based on the first predicted flow value and its corresponding label data, the loss value of the dual network deep learning algorithm model is calculated, and the model parameters are updated; the label data is the first flow value of all pipe network nodes in each secondary peak-shaving system; The dual-network deep learning algorithm model is trained in a loop until the loss value is less than a first set value, and then the training is stopped to obtain a trained dual-network deep learning algorithm model; The loss function of the dual network deep learning algorithm model during training is: Among them, L represents the loss function, q 1pi represents the first predicted flow value of the i-th pipe network node, q 1i represents the first flow value of the i-th pipe network node, and n represents the number of pipe network nodes.

7. The natural gas peak load shaving method based on time-space information according to claim 5, characterized in that: The dual-network deep learning algorithm model includes a filling network and a training network; The filling network has a first input dimension and a first output dimension; the number of dimensions of the first input dimension is numerically equal to the number of effective peak-shaving devices; the number of dimensions of the first output dimension is numerically equal to the sum of the number of all peak-shaving devices in all secondary peak-shaving systems; the effective peak-shaving devices are peak-shaving devices that work normally within a statistical period; the first input dimension is used to input the on / off status of all effective peak-shaving devices; The training network has a second input dimension and a second output dimension; the number of dimensions of the second input dimension is numerically equal to the sum of the number of all peak-shaving devices in all secondary peak-shaving systems; the number of dimensions of the second output dimension is numerically equal to the number of all pipeline nodes in the secondary peak-shaving system; the second input dimension corresponds to the first output dimension one by one and is interconnected; the result output by the second output dimension is the predicted flow value of natural gas at each pipeline node.

8. A natural gas peak load shaving method based on time-space information according to claim 7, characterized in that: The target flow value is predicted by: A10, randomly initializing the value table to obtain the Nth value table, including: randomly setting the flow value of each pipe network node in the secondary peak-shaving system as the Nth target flow value; randomly setting the adjustment amount of each pipe network node as the Nth adjustment amount; the initial value of N is set to 1; A20, calculating the Nth objective function value of each secondary peak load regulation system pipeline network node according to the Nth target flow value; The objective function is: Among them, F represents the objective function, Q represents the Nth target flow value, D represents the second demand, c represents the gas adjustment fee, cs represents the gas flow ratio in adjacent unit time, k represents the number of adjustments, and a1, a2, a3, and a4 represent the weights of each item respectively; A30, obtaining an N+1th target flow value of each pipe network node according to the Nth adjustment amount and the Nth target flow value; A40, calculating the N+1th objective function value of each secondary peak load regulation system pipeline network node according to the N+1th target flow value; A50, when the N+1th objective function value is less than the Nth objective function value, a reward value is assigned to the adjustment value corresponding to the Nth adjustment amount, and the process jumps to A60; otherwise, the process jumps directly to A60; the initial reward value of each adjustment value is 0, and the reward values ​​assigned to the same adjustment value multiple times can be accumulated; A60, updating the value table, including: selecting the N+1th adjustment amount of each pipe network node according to the set rule to obtain the N+1th value table; A70, determine whether the Nth objective function value satisfies the first set condition, if not, set N=N+1 and jump to A20, otherwise use the current Nth objective function value as the target flow value.

9. A natural gas peak load shaving method based on time-space information according to claim 8, characterized in that: The gas adjustment fee is calculated as follows: Among them, A represents the number of peak load-shaving equipment, B represents the number of gas extraction times, and H j represents the amount of natural gas regulated by the jth peak load shaving device, R j It represents the cost of peak-shaving energy per unit gas volume when the j-th peak-shaving device performs peak-shaving, U g Indicates the gas production cost per unit time, T g Indicates the gas sampling time.

10. A natural gas peak load regulation system based on time and space information, characterized in that: The system includes: The first demand prediction module is configured to obtain a second index heat map and input the trained first neural network model to predict the natural gas consumption of the target area in the first statistical period as the first demand; the second index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of the target area in the first statistical period on the geographical information of the target area; the target area includes multiple sub-areas; A delivery module is configured to deliver the first demanded amount of natural gas to the central peak-shaving system of the target area before the start of the first statistical period; the central peak-shaving system is connected to a plurality of secondary peak-shaving systems, each sub-area has a secondary peak-shaving system, and each secondary peak-shaving system includes a plurality of pipeline network nodes and a plurality of peak-shaving devices; The second demand prediction module is configured to obtain a fourth index heat map and input the trained second neural network model to predict the natural gas consumption of each sub-region of the target area within the first statistical period as the second demand; the fourth index heat map is obtained by superimposing the meteorological information, electricity consumption information and economic information of each sub-region in the target area within the first statistical period on the geographic information; The peak-shaving control module is configured to deliver each second demand amount of natural gas to the corresponding secondary peak-shaving system before the start of the first statistical period.

Images