Data fusion method and system based on multi-source data of reservoir

Abstract

The present application relates to the field of data fusion, and in particular to a data fusion method and system based on multi-source data of a reservoir. The method includes: coarsely aligning sensor data with different acquisition frequencies to obtain coarsely aligned data; training a deep learning model based on the coarsely aligned data to obtain a pre-trained finely aligned time model; setting an initial offset, inputting the coarsely aligned data with the initial offset superimposed into a branch network to obtain a feature vector, and stacking the feature vectors to obtain a two-dimensional matrix; inputting the two-dimensional matrix into the pre-trained finely aligned time model, performing secondary model training, and taking the offset parameter value in the last training result as the optimal offset; superimposing the optimal offset onto the coarsely aligned data to complete data correction, align the timestamps of the sensor data, and fusing the corrected data through a Kalman filter to obtain fused multi-source data. The present application has the effect of improving the accuracy of fused data.

Description

Technical Field

[0001] The present application relates to the field of data fusion, and in particular to a data fusion method and system based on multi-source reservoir data. Background Art

[0002] As the complexity of reservoir management systems continues to increase, various types of sensors are used in reservoirs, such as water level sensors, flow sensors, water quality monitors, weather stations, etc. These sensor networks form the cornerstone of the automated monitoring system. They collect key data such as water level, flow rate, and water quality in real time through high-precision sensors. For example, the measurement accuracy of ultrasonic water level meters is as high as ±0.2cm, ensuring the accuracy and reliability of the data. At the same time, combined with GPS positioning technology, it can achieve comprehensive coverage and real-time monitoring of all areas of the reservoir.

[0003] The fusion of sensor data is a key link in reservoir management, and one of the commonly used methods is the Kalman filter algorithm. The Kalman filter is a state filter that continuously adjusts the filter gain matrix through the Bayesian principle to reduce random interference and make the estimated system state approach the real state. In reservoir management, the Kalman filter can help fuse data from different sensors to obtain a more accurate state estimate. For example, the patent document "Spatiotemporal Dynamic Simulation and Evaluation Method of Watershed Water Ecological Products" with the authorization announcement number CN118709123B discloses determining the study area and obtaining multi-source data, preprocessing the multi-source data, forming a preprocessed data set and fusing it, and based on the fused preprocessed data set, constructing a spatiotemporal dynamic map of the watershed and a hydrological and water ecological simulation model for the study area, calculating the spatiotemporal correlation of the spatiotemporal dynamic map, and forming a spatiotemporal correlation matrix.

[0004] However, different sensors have asynchronous sampling, inconsistent data frequencies and are affected by noise, resulting in low data accuracy when using Kalman filtering for reservoir sensor data fusion, making it difficult to directly apply it to real-time monitoring and prediction tasks of reservoirs. Summary of the invention

[0005] In order to solve the problem of low accuracy of fused data, the present application provides a data fusion method and system based on multi-source data of reservoirs.

[0006] In the first aspect, the present application provides a data fusion method based on multi-source reservoir data, which adopts the following technical solutions:

[0007] The data fusion method based on multi-source data of a reservoir comprises the following steps: coarsely aligning sensor data with different acquisition frequencies to obtain coarsely aligned data; inputting the coarsely aligned data into a deep learning model, training the model once, using a mean square error loss function as the loss function, and obtaining a pre-trained finely aligned time model after the training; setting an initial offset, inputting the coarsely aligned data with the initial offset superimposed into a branch network to obtain a feature vector, stacking the feature vectors to obtain a two-dimensional matrix, wherein the number of rows in the two-dimensional matrix represents the number of sensors, and the number of columns represents the length of each sensor feature vector;

[0008] The two-dimensional matrix is ​​input into the pre-trained fine alignment time model for secondary model training. After the training, the offset parameter value in the last training result is taken as the optimal offset. The loss function of the secondary model training is: , is the loss function, represents the model output value, represents the mean value function, represents the measured data sequence processed by the Kalman filter, represents the predicted data sequence processed by the Kalman filter, Represents the output prediction value; the optimal offset is superimposed on the coarse alignment data to complete the data correction, so that the timestamps of the sensor data are aligned, and the corrected data is fused through the Kalman filter to obtain the fused multi-source data.

[0009] The beneficial effects are: the acquisition time is complemented by coarse alignment, and the data offset problem is corrected by fine alignment. The spatiotemporal synchronization of multimodal sensor data is achieved through time registration, and then data fusion and dynamic state estimation are performed through Kalman filtering, which improves the monitoring accuracy and prediction ability of reservoir hydrology, water quality and ecosystem.

[0010] Optionally, when the measured data sequence and the predicted data sequence are processed by the Kalman filter: an initial covariance matrix is ​​updated to obtain an optimal covariance matrix; and the optimal covariance matrix is ​​used to perform data fusion of the Kalman filter.

[0011] The beneficial effect is: when performing Kalman filter fusion, it is necessary to calculate the covariance matrix, because Kalman filtering will make the estimation result of Kalman filtering more biased towards data with small errors, and the covariance matrix can better reflect the linear relationship between different types of sensor data. Therefore, when time is aligned, the more positively correlated the data is, the greater its correlation should be, because the stronger the correlation of the data, the more consistent the change at the same time, but the fluctuation relationship between different sensors is not consistent, and then it is not feasible to directly use variance to evaluate the consistency of sensor data, so the optimized covariance matrix is ​​used for Kalman filter data fusion.

[0012] Optionally, the method for updating the covariance matrix includes the steps of: normalizing sensor data, calculating data differences between adjacent time points and constructing a difference sequence; constructing a two-dimensional histogram according to the difference sequence corresponding to any two sensors; in the data statistics process of the two-dimensional histogram, when the mean square error loss function value is less than a preset threshold, calculating the cumulative value of each data point of the two-dimensional histogram to obtain the frequency, and using the ratio of the average frequency to all frequencies as an update factor to update the covariance matrix; the expression of the optimal covariance matrix is: ,in, represents the optimal covariance matrix, represents the update factor, represents the initial covariance matrix.

[0013] The beneficial effects are: the ratio of the average frequency to all frequencies reflects the fluctuation consistency characteristics between different sensors. Covariance optimization is performed based on the fluctuation consistency characteristics, so that when more accurate sensor data is obtained through Kalman filtering, it can be time-synchronized to ensure the visibility of the characteristics, which is beneficial to reservoir monitoring.

[0014] Optionally, hydrological data and water quality data use neural networks to extract time series features to obtain feature vectors; meteorological data use recurrent neural networks to extract time series features to obtain feature vectors; topographic data use hyperspectral images as input, use convolutional neural networks to extract spatial features and flatten them to obtain feature vectors.

[0015] The beneficial effect is that the roughly aligned sensor data are input through different branch networks and a one-dimensional feature vector with consistent length is obtained after preprocessing.

[0016] Optionally, in normalizing the sensor data, for the image data, the image data is converted into a grayscale image, and the grayscale value of the grayscale image is normalized.

[0017] Optionally, the sensor data with different acquisition frequencies are roughly aligned, including: using the highest acquisition frequency of all sensors when acquiring data as a time reference, and performing interpolation processing on other acquired data to complete the data.

[0018] The beneficial effect is: obtain the highest acquisition frequency of all sensors when collecting data, and construct the acquisition data sequence according to the highest acquisition frequency , get the timestamp corresponding to any sensor data in the sequence. Other sensors may not have collected the corresponding data to complete the data.

[0019] Optionally, the interpolation processing method is linear interpolation, spline interpolation or piecewise constant interpolation.

[0020] In the second aspect, the present application provides a data fusion system based on multi-source data of reservoirs, which adopts the following technical solutions:

[0021] The data fusion system based on multi-source data of a reservoir comprises: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the data fusion method based on multi-source data of a reservoir is implemented.

[0022] The beneficial effect is that the above-mentioned data fusion method based on multi-source data of reservoirs is generated into a computer program and stored in a memory so as to be loaded and executed by a processor, thereby making a system based on the memory and the processor for easy use.

[0023] This application has the following technical effects:

[0024] 1. Transformer is used to time-align asynchronous data, and the adaptive Kalman filter algorithm is combined with the hierarchical method for data fusion. This solves the technical problem of time asynchrony while solving scenario problems such as asynchronous sampling of sensors, inconsistent data frequency and noise, and improves the monitoring accuracy and prediction ability of reservoir hydrology, water quality and ecosystem.

[0025] 2. Through the optimized covariance matrix and improved loss function, the time alignment network can more accurately fuse multi-source data and perform effective multi-source data fusion, thereby improving the prediction accuracy of reservoir data. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become easily understood. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-restrictive manner, and the same or corresponding numbers represent the same or corresponding parts.

[0027] Figure 1 It is a method flow chart of the data fusion method based on multi-source data of reservoirs in an embodiment of the present application. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0029] It should be understood that when the terms "first", "second", etc. are used in the claims, specification and drawings of the present application, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the specification and claims of the present application indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections.

[0030] The present application embodiment discloses a data fusion method based on multi-source data of a reservoir, which is used to improve the spatiotemporal consistency and prediction accuracy of reservoir hydrology, water quality, meteorology and ecological data, and support the scheduling, ecological management and operation and maintenance of hydraulic structures of the reservoir. Figure 1 , including steps S1 to S5, which are specifically as follows:

[0031] S1: Perform coarse alignment on sensor data with different acquisition frequencies to obtain coarse alignment data.

[0032] In one embodiment, the time series data of different sensors are collected , among which The time series data collected by each sensor is There is no restriction on the type of sensor data, which can be multi-dimensional data of the same type, single-dimensional data of the same type, or other types of data such as image data. That is, X is a time series data, which may be one-dimensional data, multi-dimensional data, or image data.

[0033] Due to the inconsistency of the collected data dimensions, the collection frequency of different data is inconsistent. For example: water level data sensors collect data every 15 minutes, rainfall data sensors collect data every 10 minutes, flow data sensors collect data every 5 minutes, evaporation data sensors collect data every 12 hours, dissolved oxygen, pH, conductivity, turbidity and other conventional water quality sensors collect data every hour, and other types of sensors used to collect related data are not described in detail. The type of collection sensor can be adjusted by the implementer according to the specific implementation scenario.

[0034] The acquisition frequency and asynchronous characteristics of various sensor data significantly affect the complexity of data fusion. The sampling frequencies of different data sources vary greatly, resulting in the inability of sensors to record synchronously in time. The inconsistent sampling frequencies of different data sources make it impossible to align the data directly. This asynchronous characteristic is the main challenge of data fusion, especially in real-time applications. Effectively handling these time mismatches is the key technical difficulty in realizing multi-source data fusion. In addition, some sensors may be greatly disturbed during the acquisition process, resulting in unstable acquisition results, and thus data fusion (Kalman filter data fusion) methods are needed to obtain more accurate data.

[0035] The method of rough alignment is: obtain the highest acquisition frequency of all sensors when collecting data, and construct the acquisition data sequence according to the highest acquisition frequency , get the timestamp corresponding to any sensor data in the sequence. Other sensors may not have collected the corresponding data. Then, use the following method to interpolate the collected data of other sensors to obtain the data of other sensors at the timestamp.

[0036] Use the highest acquisition frequency of all sensors when collecting data as the time reference, and interpolate other collected data to complete the data. Corresponding timestamps are used to obtain data from other sensors at the timestamps. However, since other sensors may not have collected the corresponding data, the collected data from other sensors are interpolated using the following method to complete the data.

[0037] The interpolation processing method can be linear interpolation, spline interpolation or piecewise constant interpolation. In one embodiment, different types of data can use different interpolation methods according to their characteristics. For example, for relatively high-frequency data such as hydrological data, linear interpolation or spline interpolation is used, and relatively low-frequency data such as evaporation uses piecewise constant interpolation.

[0038] After interpolation, we get The result after rough alignment of the sensors is , after getting all sensors After that, since it needs to be input into the branch network, the position encoding method is used to encode the data of each sensor. Position encoding is performed, where the position encoded data is The position encoding process is a prior art, and thus this solution will not be described in detail.

[0039] The data corresponding to each sensor after rough alignment The time series length is kept consistent within the same period of time, and the data of each sensor after rough alignment is recorded as . Use position encoding to encode each Position encoding is performed for subsequent precise alignment.

[0040] However, the above coarse alignment only makes the timestamps consistent (that is, the above coarse alignment process only makes the time resolution consistent and fills it in according to the acquisition time). Since the time of the acquisition sensor may be affected by environmental factors and other interferences, there may be errors, that is, the timestamps of different sensors are not unified, that is, common hardware delays, communication delays, clock drift, and nonlinear system behavior problems, and then fine alignment is performed.

[0041] S2: Input the coarse alignment data into the deep learning model and train the model once. The loss function is the mean square error loss function. After the training, the pre-trained fine alignment time model is obtained.

[0042] The deep learning model is the Transformer model, and the mean square error loss function is an existing technology, which will not be described in detail here. The mean square error loss function is used to pre-train a data prediction regression model. If a data prediction regression model is not pre-trained, it is impossible to determine whether the time alignment network should adjust the time offset or adjust the network parameters to ensure the prediction regression effect.

[0043] When the mean square error loss function is equal to the threshold of 0.5, it means that the pre-training model training is completed. 0.5 is a hyperparameter and can be adjusted according to the actual application scenario.

[0044] S3: Set the initial offset, input the coarse alignment data with the superimposed initial offset into the branch network to obtain a feature vector, and stack the feature vectors to obtain a two-dimensional matrix. The number of rows in the two-dimensional matrix represents the number of sensors, and the number of columns represents the length of each sensor feature vector.

[0045] The roughly aligned sensor data are input through different branch networks and preprocessed to obtain a one-dimensional feature vector with consistent length, which is a column vector with a length of m.

[0046] In one embodiment, each branch network is an existing network structure. Hydrological data and water quality data use NN (Neural Network) to extract time series features and output 1×m feature vectors. Meteorological data uses RNN (Recurrent Neural Network) to extract time series features and output 1×m feature vectors. Topographic data uses hyperspectral images as input, uses CNN (Convolutional Neural Networks) to extract spatial features, and flattens them into 3×m vectors.

[0047] The sensor data sequences are preprocessed and stacked in columns to obtain a dimension of A two-dimensional matrix, where is the sum of the number of dimensions of each sensor and the number of dimensions of all image data after flattening. The two-dimensional matrix is ​​used to input into Transformer for fine-grained temporal alignment, that is: As input data to the time-aligned Transformer network.

[0048] Since each sensor data has its own acquisition time offset, the goal of the model is to eliminate these offsets so that the output data is aligned on a unified time basis. That is, a time offset is added to each of the above sensor data when it is input into the time alignment Transformer network. , No. The offset of the sensor data is The superposition offset refers to the forward or backward offset of the input data when selecting data. The amount of data corresponding to time.

[0049] S4: Input the two-dimensional matrix into the pre-trained fine alignment time model, perform secondary model training, and after the training, take the offset parameter value in the last training result as the optimal offset.

[0050] Overlaying the offset As the input of the time-aligned Transformer, the time-aligned Transformer network result is the existing technology, and implementers can choose other networks that can perform fine alignment according to the specific implementation scenario.

[0051] Among them, the loss function of the secondary model training is:

[0052] , is the loss function, represents the model output value, represents the mean value function, represents the measured data sequence processed by the Kalman filter, represents the predicted data sequence processed by the Kalman filter, Represents the output prediction value.

[0053] In one embodiment, the training is completed when the loss function value of the secondary training is less than 0.01 or the number of training times reaches 1000 times.

[0054] During the training process, It should be close In the data fusion process of the Kalman filter, the Kalman filter itself can continuously make predictions, estimates and updates. In the training set of the time alignment network, the data equivalent to the current future moment is known. Through 10 consecutive data fusion processes of the Kalman filter, we get and , where 10 is a hyperparameter that can be adjusted by the implementer according to the specific implementation scenario.

[0055] The final time offset ensures that the time is aligned as much as possible and can obtain more accurate reservoir sensor data for reservoir monitoring when performing data fusion reasoning in the future.

[0056] In one embodiment, the method for processing the measured data sequence and the predicted data sequence in the Kalman filter is to update the initial covariance matrix to obtain the optimal covariance matrix; and use the optimal covariance matrix to perform data fusion of the Kalman filter.

[0057] When performing Kalman filter fusion, it is necessary to calculate the covariance matrix, because Kalman filtering will make the estimation results of Kalman filtering more biased towards data with small errors. The covariance matrix can better reflect the linear relationship between different types of sensor data. Therefore, when time is aligned, the more positively correlated the data is, the greater the correlation should be. Because the stronger the correlation of the data, the more consistent the change at the same time. However, the fluctuation relationship between different sensors is not consistent. Therefore, it is not feasible to directly use variance to evaluate the consistency of sensor data. Then, the fluctuation consistency characteristics between different sensors are obtained, and the covariance is optimized according to the fluctuation consistency characteristics, so that when more accurate sensor data is obtained through Kalman filtering, it can be time synchronized to ensure the obviousness of the characteristics, which is beneficial to reservoir monitoring.

[0058] In one embodiment, the method of updating the covariance matrix comprises the steps of:

[0059] Normalize the sensor data, calculate the data differences between adjacent time points and construct a difference sequence.

[0060] If it is image data, it is converted into a grayscale image and grayscale value normalization is performed, or it is converted into other single-dimensional images and then normalized according to the specific implementation scenario. The length of the data collected by the two sensors is 20 data, and 20 is a hyperparameter that can be adjusted by the implementer according to the specific implementation scenario.

[0061] After normalization, the difference sequence between the normalized data corresponding to a single sensor and the normalized data at the previous moment is obtained. The difference sequence corresponding to another sensor When making a difference, make the difference directly without adding the absolute value.

[0062] A two-dimensional histogram is constructed according to the difference sequence corresponding to any two sensors. In the data statistics process of the two-dimensional histogram, when the mean square error loss function value is less than the preset threshold, the cumulative value of each data point in the two-dimensional histogram is calculated to obtain the frequency (frequency), and the ratio of the average frequency to all frequencies is used as the update factor to update the covariance matrix. The update factor is used as a feature that reflects the fluctuation consistency of the data collected by the two sensors.

[0063] Frequency can be obtained through two-dimensional histogram statistics. Average frequency refers to the average frequency of a single data point at multiple times. All frequencies refer to the total number of all data points. The number of data points is the same at different times.

[0064] The expression of the optimal covariance matrix is: ,in, represents the optimal covariance matrix, represents the update factor, represents the initial covariance matrix. The bigger, The smaller it is, the smaller the variance between sensors with higher correlation after updating. Therefore, when the data of all sensors are aligned, they tend to be more correlated sensors. Therefore, after the data of different sensors are aligned, the correlation is strengthened, which is more conducive to abnormal monitoring of the reservoir.

[0065] Through the optimized covariance matrix and improved loss function, the time alignment network can more accurately fuse multi-source data and perform effective multi-source data fusion, thereby improving the prediction accuracy of reservoir data.

[0066] S5: Superimpose the optimal offset to the coarse alignment data to complete data correction, align the timestamps of the sensor data, and fuse the corrected data through a Kalman filter to obtain fused multi-source data.

[0067] An embodiment of the present application also discloses a data fusion system based on multi-source data of a reservoir, including a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, a data fusion method based on multi-source data of a reservoir according to the present application is implemented.

[0068] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, and their configuration and functions are known in the art, so they will not be described in detail here.

[0069] In the present application, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high bandwidth memory HBM (High Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device.

[0070] Although this specification has shown and described a plurality of embodiments of the present application, it is obvious to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will think of many changes, modifications and alternatives without departing from the thought and spirit of the present application. It should be understood that in the process of practicing the present application, various alternatives to the embodiments of the present application described herein may be adopted.

[0071] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A data fusion method based on multi-source reservoir data, characterized in that: Includes steps: Performing rough alignment on sensor data with different acquisition frequencies to obtain rough alignment data; The coarse alignment data is input into the deep learning model and the model is trained once. The loss function is the mean square error loss function. After the training, the pre-trained fine alignment time model is obtained. Set the initial offset, input the coarse alignment data superimposed with the initial offset into the branch network to obtain a feature vector, stack the feature vectors to obtain a two-dimensional matrix, the number of rows in the two-dimensional matrix represents the number of sensors, and the number of columns represents the length of each sensor feature vector; Input the two-dimensional matrix into the pre-trained fine alignment time model, perform secondary model training, and after the training, take the offset parameter value in the last training result as the optimal offset; Among them, the loss function of the secondary model training is: , is the loss function, represents the model output value, represents the mean value function, represents the measured data sequence processed by the Kalman filter, represents the predicted data sequence processed by the Kalman filter, Represents the output prediction value; The optimal offset is superimposed on the coarse alignment data to complete the data correction, so that the timestamps of the sensor data are aligned, and the corrected data is fused through the Kalman filter to obtain the fused multi-source data.

2. The data fusion method based on multi-source reservoir data according to claim 1 is characterized in that: When the Kalman filter processes the measured data sequence and the predicted data sequence: the initial covariance matrix is ​​updated to obtain the optimal covariance matrix; the optimal covariance matrix is ​​used to perform data fusion of the Kalman filter.

3. The data fusion method based on multi-source data of reservoir according to claim 2 is characterized in that: The method for updating the covariance matrix comprises the steps: Normalize sensor data, calculate data differences between adjacent time points and construct a difference sequence; Construct a two-dimensional histogram based on the difference sequence corresponding to any two sensors; In the data statistics process of the two-dimensional histogram, when the mean square error loss function value is less than the preset threshold, the cumulative value of each data point of the two-dimensional histogram is calculated to obtain the frequency, and the ratio of the average frequency to all frequencies is used as an update factor to update the covariance matrix; The expression of the optimal covariance matrix is: ,in, represents the optimal covariance matrix, represents the update factor, represents the initial covariance matrix.

4. The data fusion method based on multi-source data of a reservoir according to claim 3 is characterized in that: For hydrological data and water quality data, neural networks are used to extract time series features to obtain feature vectors; for meteorological data, recurrent neural networks are used to extract time series features to obtain feature vectors; for topographic data, hyperspectral images are used as input, and convolutional neural networks are used to extract spatial features and flatten them to obtain feature vectors.

5. The data fusion method based on multi-source data of reservoir according to claim 4 is characterized in that: In normalizing sensor data, for image data, the image data is converted into a grayscale image, and the grayscale value of the grayscale image is normalized.

6. The data fusion method based on multi-source reservoir data according to claim 1 is characterized in that: The sensor data with different acquisition frequencies are roughly aligned, including: using the highest acquisition frequency of all sensors when collecting data as a time reference, and interpolating other collected data to complete the data.

7. The data fusion method based on multi-source data of reservoir according to claim 6 is characterized in that: The interpolation processing method is linear interpolation, spline interpolation or piecewise constant interpolation.

8. The data fusion system based on multi-source data of reservoirs is characterized by: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the data fusion method based on multi-source data of a reservoir according to any one of claims 1 to 7 is implemented.

Images