An intelligent identification and early warning method, device and storage medium for water supply network system leakage events

Abstract

The present invention discloses a method, device and storage medium for intelligent identification and early warning of leakage events in a water supply network system, which relates to the technical field of water supply network control. Among them, the method includes: obtaining flow monitoring data and performing preprocessing; inputting the feature vector into the trained prediction model to output the predicted value at time t + 1; performing feedback correction on the predicted value output by the prediction model, and determining whether to adjust the input of the prediction model at time t + 2 according to the deviation between the predicted value and the measured value; using the Pauta criterion to determine whether the number of times the residual between the predicted value and the measured value of a single day exceeds a specified multiple of the standard deviation exceeds a preset value: if so, it is identified that a leakage event has occurred and an alarm is issued; otherwise, no alarm operation is performed. Compared with the prior art, the present invention does not need to input a large amount of historical flow detection data into the model, and avoids the error caused by the input of abnormal data through feedback correction, realizes the long-term stable use of the prediction model, and improves the early warning accuracy of leakage events.

Description

Technical Field

[0001] The present invention relates to the technical field of water supply network control, and more specifically, to an intelligent identification and early warning method, device and storage medium for leakage events in a water supply network system. Background Art

[0002] The water supply network is a main component of the urban lifeline project, playing a role in maintaining the normal operation and economic function of the city. When a leakage occurs in the water supply network, the data collected by the flow sensors near the leakage point will be abnormal. Therefore, a prediction-classification method in data-driven can be used for anomaly detection to achieve intelligent identification and early warning of leakage events.

[0003] In the traditional prediction-classification method, the model training process in the prediction stage often inputs more samples into the model so that the model can maintain high accuracy and availability in the future. However, in the domestic water supply network system, the data acquisition and monitoring control system was built relatively late and may not be able to provide a large amount of historical monitoring data for 3 - 5 years. Even if the water supply companies in some developed cities may be able to provide this data, they still face the following problems: (1) China's cities are developing rapidly, with large population changes. In the past 3 - 5 years, the topological structure of the water supply network and the user water consumption pattern may have changed greatly. For example, the original industrial area may have become a residential area, etc. This series of changes may lead to a low correlation between historical data and future data and lack of training value; (2) Inputting a large dataset for 3 - 5 years at one time will greatly increase the difficulty of model training and require too high computer computing power.

[0004] In the traditional prediction-classification method, the model prediction process in the prediction stage often uses the traditional iterative prediction method. In the actual use process, the error of the traditional iterative prediction method will accumulate continuously along the time axis, thus causing a large error. In the classification stage of the traditional prediction-classification method, single-threshold classification is mostly used. The data collected by the flow sensors in the water supply network are affected by the user water consumption fluctuation and will show large fluctuations. Moreover, valve scheduling in the water supply network, quality problems of flow sensors, etc. will all cause abnormal flow monitoring data. Therefore, single-threshold classification is difficult to identify accurate leakage signals. Summary of the Invention

[0005] The present invention aims to overcome the above-mentioned defects of the prior art, such as the need for a large amount of historical data and poor accuracy, and provides an intelligent identification and early warning method, device and storage medium for leakage events in a water supply network system.

[0006] To solve the above technical problems, the technical solution of the present invention is as follows:

[0007] In a first aspect, an intelligent identification and early warning method for leakage events in a water supply network system includes:

[0008] Obtain flow monitoring data and perform preprocessing; the flow monitoring data includes historical flow measured values and current flow measured values at time t;

[0009] Generate feature vectors based on the preprocessed flow monitoring data;

[0010] Input the feature vectors into the trained prediction model to output the predicted value at time t+1; the prediction model includes an input layer, a hidden layer, and an output layer, and the hidden layer includes a GRU layer, a Dropout layer, and an Attention layer;

[0011] Perform feedback correction on the predicted value output by the prediction model, and confirm whether to adjust the input of the prediction model at time t+2 according to the deviation between the predicted value and the measured value;

[0012] Adopt the Pauta criterion to determine whether the number of times the residual between the predicted value and the measured value on a single day exceeds a specified multiple of the standard deviation exceeds a preset value: if so, identify that a leakage event has occurred and issue an alarm; otherwise, do not perform the alarm operation.

[0013] In a second aspect, a computer device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0014] In a third aspect, a computer storage medium stores instructions, and when the instructions are executed on a computer, the computer is made to execute the method described in the first aspect.

[0015] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: the present invention does not require a large amount of historical flow detection data as input to train the model, and through feedback correction, it avoids the errors caused by the input of abnormal data, realizing the long-term stable use of the prediction model; in addition, the present invention reduces false alarms caused by single outliers, sensor anomalies, and valve scheduling, etc., and improves the early warning accuracy of leakage events. Description of the Drawings

[0016] Figure 1 It is a schematic flow chart of the intelligent identification and early warning method for leakage events in the water supply network system in Embodiment 1;

[0017] Figure 2 It is a schematic flow chart of feedback correction in Embodiment 1;

[0018] Figure 3 It is a schematic topological structure diagram of the leakage location and the flow sensor location in Embodiment 2;

[0019] Figure 4 It is a schematic diagram of historical flow monitoring data in Embodiment 2;

[0020] Figure 5 Schematic diagram of the filling and masking process in Embodiment 2;

[0021] Figure 6 Schematic diagram of the prediction model structure in Embodiment 2;

[0022] Figure 7 Comparison chart of predicted values and measured values of traffic data on November 20, 2021 in Embodiment 2;

[0023] Figure 8 Comparison chart of predicted values and measured values of traffic data on November 21, 2021 in Embodiment 2;

[0024] Figure 9 Comparison chart of predicted values and measured values of traffic data on November 22, 2021 in Embodiment 2;

[0025] Figure 10 Comparison chart of predicted values and measured values of traffic data on November 23, 2021 in Embodiment 2;

[0026] Figure 11 Comparison chart of predicted values and measured values of traffic data on November 24, 2021 in Embodiment 2. Specific implementation mode

[0027] The attached drawings are only for illustrative purposes and should not be construed as a limitation of this patent;

[0028] For better illustration of this embodiment, some components in the attached drawings will be omitted, enlarged or reduced, which do not represent the dimensions of the actual product;

[0029] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the attached drawings may be omitted.

[0030] The technical solutions of the present invention will be further described below with reference to the attached drawings and embodiments.

[0031] Embodiment 1

[0032] This embodiment proposes an intelligent identification and early warning method for water leakage events in a water supply network system. Refer to Figure 1 , including:

[0033] Obtain traffic monitoring data and perform preprocessing; the traffic monitoring data includes historical measured traffic values and the measured traffic value at the current time t;

[0034] Generate feature vectors based on the preprocessed traffic monitoring data;

[0035] Input the feature vector into the trained prediction model to output the predicted value at time t+1; the prediction model includes an input layer, a hidden layer, and an output layer, and the hidden layer includes a GRU layer, a Dropout layer, and an Attention layer;

[0036] Perform feedback correction on the predicted value output by the prediction model, and confirm whether to adjust the input of the prediction model at time t+2 according to the deviation between the predicted value and the measured value;

[0037] Adopt the Pauta criterion to determine whether the number of times the residual between the predicted value and the measured value on a single day exceeds a specified multiple of the standard deviation exceeds a preset value: if so, identify that a leakage event has occurred and issue an alarm; otherwise, do not perform the alarm operation.

[0038] In the existing iterative prediction scheme, the flow prediction data at the current moment is continuously input to predict the flow data at the next moment, so as to achieve rolling prediction. Although this scheme is simple and fast, there is an obvious problem, that is, each time the predicted flow data is used for the next-step prediction, the prediction error will accumulate continuously along the time axis, resulting in a large error. In this embodiment, an improved multi-step prediction method is proposed, where the predicted flow data is no longer input, but the measured flow data at the current moment is input for the next-step prediction, and only in the feedback correction link, it is confirmed whether it is necessary to adjust the input of the prediction model at time t+2 according to the deviation between the predicted value and the measured value, improving the prediction accuracy.

[0039] In a preferred embodiment, the expression of the standard deviation is as follows:

[0040]

[0041] In the formula, σ t represents the standard deviation at time t; x tj represents the measured value at time t on the jth day; represents the average value of the measured data of the flow data at time t for all days used for model training, and its expression is as follows:

[0042]

[0043] In the formula, n represents all days of the flow data used for model training.

[0044] In a preferred embodiment, the preprocessing includes:

[0045] For the duplicate values in the flow monitoring data, keep one of them and delete the others;

[0046] For the missing values in the flow monitoring data, use the linear interpolation method to fill them, and the interpolation process is shown in the following formula:

[0047]

[0048] Among them, x t represents the value to be filled at time t, that is, the missing value; x t+1 is the measured value at time t + 1, that is, the moment after the missing value; x t-1 is the measured value at time t - 1, that is, the moment before the missing value; Δt represents the time difference between the two moments of time t + 1 and t - 1;

[0049] For the inconsistent acquisition frequencies of flow monitoring data, resampling the time series and using linear interpolation are adopted to transform the time series into the expected frequency;

[0050] Normalize the flow monitoring data, and the normalization process is shown in the following formula:

[0051]

[0052] Among them, X min is the minimum value in the input sequence data; X max is the maximum value in the input sequence data; X i is the original value in the input sequence data; X i ' is the value after normalizing the original data in the input sequence data.

[0053] Resampling refers to the conversion of a time series from one time step to another.

[0054] In a specific implementation process, for the problem of inconsistent acquisition frequencies, first upsample the time series, that is, encrypt it into a high-precision time series with consistent frequencies, then use the linear filling method to fill the flow values corresponding to the newly added time series, and finally process the time series into the desired frequency through downsampling.

[0055] In a preferred embodiment, generating the feature vector based on the preprocessed flow monitoring data includes:

[0056] Extract the flow monitoring data of the most recent 30 days from the preprocessed flow monitoring data to generate 23×k samples. The samples consider the periodicity of each week and the trend of each day, and include two groups of feature vectors, as shown below:

[0057]

[0058] Among them, x represents the measured monitoring data, and k represents the number of data collected in one day;

[0059] Introduce a filling mechanism to supplement the value "-1" as a placeholder into the feature vectors with shorter lengths, so that the lengths of the feature vectors in the samples are consistent, and introduce a masking mechanism to block the placeholders.

[0060] In this preferred embodiment, features and labels are input into the prediction model in pairs to achieve supervised learning, where the label represents the predicted value based on the corresponding moment of a single day, and the predicted value is generated based on the k traffic monitoring data collected before the corresponding moment and the traffic monitoring data collected at the same moment in the previous 7 days.

[0061] In addition, in this preferred embodiment, since the lengths of the feature vectors of the two sets of data in the sample are inconsistent, while GRU requires that the input samples have the same length. When multiple samples with different lengths are input, a padding mechanism and a masking mechanism are added to ensure the normal operation of the model. Since there are no negative values in the historical traffic monitoring data, -1 is used as a placeholder to process two feature vectors with different lengths into vectors of equal length, and then the -1 is masked through the masking mechanism to ensure that the introduced placeholder does not affect the model accuracy.

[0062] In a preferred embodiment, in the hidden layer of the prediction model, the GRU layer includes a first GRU layer and a second GRU layer, and the dropout layer includes a first dropout layer and a second dropout layer; wherein, the first GRU layer, the first dropout layer, the second GRU layer, the second dropout layer, and the Attention layer are connected in sequence.

[0063] In this preferred embodiment, a Dropout layer is added after each GRU layer to randomly inactivate some neurons, which can prevent the prediction model from overfitting and enhance the generalization ability of the model.

[0064] In a preferred embodiment, refer to Figure 2 , x t represents the measured value of the traffic at time t; y tt represents the predicted value of the traffic at time t. The predicted value output by the prediction model is feedback-corrected, and it is confirmed whether to adjust the input of the prediction model at time t + 2 according to the deviation between the predicted value and the measured value. Specifically:

[0065] If the deviation between the predicted value at time t + 1 output by the prediction model and the measured value at time t + 1 obtained exceeds 1 standard deviation, then at time t + 2, the input of the prediction model includes the measured value of the historical traffic and the predicted value at time t + 1, and does not include the measured value at time t + 1.

[0066] In a preferred embodiment, the method of using the Pauta criterion to determine whether the number of times the residual between the predicted value and the measured value of a single day exceeds a specified multiple of the standard deviation exceeds a preset value includes:

[0067] (1) The moments when the residual between the predicted value and the measured value of a single day exceeds σ t are marked as 1, and an alarm is given when 8 consecutive marked moments are greater than or equal to 1;

[0068] (2) Mark the time when the residual between the single-day predicted value and the measured value exceeds 2σ as 2, and issue an alarm when 4 consecutive time marks are greater than or equal to 2; t

[0069] (3) Mark the time when the residual between the single-day predicted value and the measured value exceeds 4σ as 4, and issue an alarm when 2 consecutive time marks are greater than or equal to 4; t

[0070] (4) Mark the time when the residual between the single-day predicted value and the measured value exceeds 8σ as 8, and issue an alarm when 1 consecutive time mark is greater than or equal to 8. t

[0071] In an optional embodiment, the prediction model is trained using a feedback correction training method, including:

[0072] Use the enumeration method to list various combinations and construct an initial prediction model with different numbers of neurons and related parameters;

[0073] Write the flow monitoring data under normal conditions for 30 days into a 30×k feature matrix as follows:

[0074]

[0075] where x represents the measured monitoring data and k represents the number of data collected by the sensor in one day;

[0076] Input the feature matrix into the prediction model for initial training, calculate the mean square error MSE based on the predicted value and the measured value of the 31st day output by the prediction model, and retain the prediction model parameters when the mean square error MSE is the smallest as the optimal parameters. The expression of the mean square error MSE is as follows:

[0077]

[0078] where x i is the measured value, is the predicted value output by the prediction model, and n is the number of data;

[0079] Use the Pauta criterion to classify the predicted value of the 31st day. When the classification result shows normal, directly retrain the model with the measured values from the 2nd day to the 31st day and then predict the data of the 32nd day; when the classification result shows abnormal, retrain the model with the measured values from the 2nd day to the 30th day and the predicted value of the 31st day and then predict the data of the 32nd day.

[0080] Since the flow monitoring data in the water supply pipe network is affected by the daily and nightly changes in user water consumption, the daily monitoring data will show severe fluctuations with peaks and valleys. In this preferred embodiment, an anomaly detection method that changes with the date is adopted, improving the adaptability of the detection results to fluctuations.

[0081] In a specific implementation process, when the classification result shows an anomaly, it indicates that a pipe burst has occurred in the water supply pipe network system.

[0082] It can be understood that the model continuously trains and predicts the subsequent days according to this method. When the model performs well, it can be used continuously. When the model accuracy error is large, the model structure is appropriately adjusted and then used.

[0083] Embodiment 2

[0084] To verify the feasibility of the method described in Embodiment 1, in this embodiment, a simulation experiment was conducted on the method described in Embodiment 1 using the leakage event data on a DN500 pipe section of a water supply pipe network system.

[0085] Figure 3 Shows the topological structure of the leakage location and the flow sensor location on this pipe section; Figure 4 Shows the flow monitoring data of this pipe section from October 20, 2021 to November 24, 2021. It is known that the actual repair time recorded by the water company staff is 13:00 on November 22, 2021, and the flow sensor records the flow data every 5 minutes.

[0086] Obtain the flow monitoring data from October 20, 2021 to November 23, 2021 and perform preprocessing. Some of the preprocessed flow monitoring data is shown in Table 1:

[0087] Table 1 Preprocessed flow monitoring data

[0088]

[0089]

[0090] Extract features from the preprocessed flow monitoring data, including:

[0091] (1) Write the flow monitoring data of 30 days in normal operating conditions from 00:00 on October 20, 2021 to 23:55 on November 18, 2021 into a 30×288 feature matrix as follows:

[0092]

[0093] (2) Considering the weekly periodicity and daily trend, generate 6624 samples as follows:

[0094]

[0095] Refer to Figure 5 , introduce a padding mechanism to supplement the value "-1" as a placeholder into the feature vectors with shorter lengths, so that the lengths of the feature vectors in the samples are kept consistent, and introduce a masking mechanism to mask the placeholder.

[0096] Since there are two feature vectors in the sample, the hidden layer of the prediction model adopts a structure connected by 2 GRU layers, 2 dropout layers and 1 Attention layer. Refer to Figure 6 . Among them, the number of neurons and other related parameters are searched by the enumeration method to construct multiple initial prediction models.

[0097] Input the feature matrix corresponding to the 30-day traffic monitoring data from 2021-10-20 00:00 to 2021-11-18 23:55 into multiple initial prediction models for initial training, and predict the traffic on 2021-11-19. Use the mean square error to evaluate the accuracy of multiple initial models. The evaluation results are shown in Table 2:

[0098] Table 2 Comparison list of mean square error results of different prediction models

[0099]

[0100]

[0101] It can be seen that the optimal parameter combination is the parameter of the prediction model corresponding to Experiment No. 1. Use the prediction model corresponding to this group of parameters for subsequent prediction experiments.

[0102] Use the model corresponding to the above optimal parameter combination to predict the traffic data on 2021-11-20. Figure 7 Shows the comparison between the predicted results and the measured results of the traffic data on 2021-11-20. It can be seen that the day is in normal working condition and no leakage is detected. The predicted results fit the measured results.

[0103] Use this prediction model to continue predicting the traffic data on 2021-11-21. Since 2021-11-20 is in normal working condition, directly replace the data on 2021-10-21 with the measured data on 2021-11-20 and input it into the prediction model. Figure 8 Shows the comparison between the predicted results and the measured results of the traffic data on 2021-11-20. It can be seen that after 03:05, the residual between the measured value and the predicted value output by the prediction model exceeds 2 times the standard deviation. Based on the Pauta criterion, the prediction model issues a leakage alarm at 03:20, 4 moments later, which is 33 hours and 40 minutes earlier than the water company staff, avoiding a large amount of water resource waste.

[0104] Using this prediction model, continue to predict the traffic data on November 22, 2021. Since the leakage condition occurred on November 21, 2021, through feedback correction, adjust the input of the prediction model, and replace the monitoring data on October 22, 2021 with the predicted value on November 21, 2021 and input it into the prediction model. Figure 9 The comparison between the predicted results and the measured results of the traffic data on November 22, 2021 is shown. The prediction model identifies a leakage event and issues an alarm, that is, a continuous leakage condition is identified.

[0105] Using this prediction model, continue to predict the traffic data on November 23, 2021. Since the leakage condition occurred on November 22, 2021, through feedback correction, adjust the input of the prediction model, and replace the monitoring data on October 23, 2021 with the predicted value on November 22, 2021 and input it into the prediction model. Figure 10 The comparison between the predicted results and the measured results of the traffic data on November 23, 2021 is shown. The prediction model identifies a continuous leakage condition.

[0106] Using this prediction model, continue to predict the traffic data on November 24, 2021. Since the leakage condition occurred on November 22, 2021, through feedback correction, adjust the input of the prediction model, and replace the monitoring data on October 24, 2021 with the predicted value on November 23, 2021 and input it into the prediction model. Figure 11 The comparison between the predicted results and the measured results of the traffic data on November 24, 2021 is shown. The predicted value output by the prediction model after 01:30 shows that the official website has returned to normal, and the residual between the predicted value and the measured value is within a reasonable deviation range.

[0107] Embodiment 3

[0108] This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the method described in Embodiment 1 is implemented.

[0109] Embodiment 4

[0110] This embodiment provides a computer storage medium, characterized in that the computer storage medium stores instructions, and when the instructions are executed on a computer, the computer executes the method described in Embodiment 1.

[0111] The same or similar reference numerals correspond to the same or similar components;

[0112] The terms describing the positional relationship in the drawings are only for illustrative purposes and should not be construed as a limitation of this patent;

[0113] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all implementation manners here. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. An intelligent identification and early warning method for water supply network system leakage events, characterized in that, Including: Obtain traffic monitoring data and perform preprocessing; The traffic monitoring data includes historical traffic measured values and the traffic measured value at the current moment t; Generate feature vectors based on the preprocessed traffic monitoring data; Input the feature vectors into the trained prediction model to output the predicted value at the moment t+1; the prediction model includes an input layer, a hidden layer, and an output layer, and the hidden layer includes a GRU layer, a Dropout layer, and an Attention layer; Perform feedback correction on the predicted value output by the prediction model, and confirm whether to adjust the input of the prediction model at the moment t+2 according to the deviation between the predicted value and the measured value; Adopt the Pauta criterion to determine whether the number of times the residual between the predicted value and the measured value on a single day exceeds a specified multiple of the standard deviation exceeds a preset value: if so, identify that a leakage event has occurred and issue an alarm; Otherwise, do not perform the alarm operation; Among them, the preprocessing includes: For the duplicate values in the traffic monitoring data, keep one of them and delete the rest; For the missing values in the traffic monitoring data, use the linear interpolation method to fill them, and the interpolation process is shown in the following formula: Among them, x t represents the value to be filled at time t, that is, the missing value; x t+1 is the measured value at time t + 1, that is, the moment after the missing value; x t-1 is the measured value at time t - 1, that is, the moment before the missing value; Δt represents the time difference between the two moments of time t + 1 and t - 1; For the inconsistent acquisition frequencies of the traffic monitoring data, use resampling and linear interpolation of the time series to convert the time series into the expected frequency; Normalize the traffic monitoring data, and the normalization process is shown in the following formula: Among them, X min is the minimum value in the input sequence data; X max is the maximum value in the input sequence data; X i is the original value in the input sequence data; X i ' is the value after normalizing the original data in the input sequence data; and, The expression of the standard deviation is as follows: Where, σ t represents the standard deviation at time t; x tj represents the measured value at time t on the j-th day; represents the average value of the measured data of the flow data at time t for all days used for model training, and its expression is as follows: In the formula, n represents the total number of days of traffic data used for model training; and, Generating feature vectors based on the preprocessed traffic monitoring data includes: Extract the traffic monitoring data of the most recent 30 days from the preprocessed traffic monitoring data to generate 23×k samples. The samples consider the weekly periodicity and daily trend, and include two groups of feature vectors, as follows: Among them, x represents the measured monitoring data, k represents the number of data collected in one day; the label represents the predicted value based on the corresponding moment of a single day, and this predicted value is generated based on the k traffic monitoring data collected before the corresponding moment and the traffic monitoring data collected at the same moment in the previous 7 days; Introduce a padding mechanism to supplement the value "-1" as a placeholder into the feature vector with a shorter length to make the length of the feature vectors in the sample consistent, and introduce a masking mechanism to mask the placeholder; and, Adopting the Pauta criterion to determine whether the number of times the residual between the predicted value and the measured value on a single day exceeds a specified multiple of the standard deviation exceeds a preset value includes: (1) The moment when the residual between the single-day predicted value and the measured value exceeds σ t is marked as 1, and an alarm is triggered when 8 consecutive moments are marked as greater than or equal to 1. (2) When the residual between the single-day predicted value and the measured value exceeds 2σ t the time mark is marked as 2, and an alarm is issued when 4 consecutive time marks are greater than or equal to 2; (3) The moments when the residuals between the single-day predicted values and the measured values exceed 4σ t are marked as 4, and an alarm is triggered when two consecutive moments are greater than or equal to 4. (4) The moment when the residual between the single-day predicted value and the measured value exceeds 8σ t is marked as 8, and an alarm is issued when 1 consecutive moment is greater than or equal to 8.

2. The intelligent identification and early warning method for water supply network system leakage events according to claim 1, wherein, In the hidden layer of the prediction model, the GRU layer includes a first GRU layer and a second GRU layer, and the Dropout layer includes a first dropout layer and a second dropout layer; among them, the first GRU layer, the first dropout layer, the second GRU layer, the second dropout layer, and the Attention layer are connected in sequence.

3. The intelligent identification and early warning method for water supply network system leakage events according to claim 1, characterized in that Performing feedback correction on the predicted value output by the prediction model, and confirming whether to adjust the input of the prediction model at the moment t+2 according to the deviation between the predicted value and the measured value is specifically: If the deviation between the predicted value at time t+1 output by the prediction model and the measured value at time t+1 obtained exceeds 1 standard deviation, then at time t+2, the input of the prediction model includes the measured value of the historical flow rate and the predicted value at time t+1, and does not include the measured value at time t+1.

4. The intelligent identification and early warning method for water supply network system leakage events according to claim 1, characterized in that The prediction model is trained using a feedback correction training method, including: Using the enumeration method to list various combinations to construct an initial prediction model with different numbers of neurons and related parameters; Write the flow rate monitoring data under normal conditions for 30 days into a 30×k feature matrix as follows: Among them, x represents the measured monitoring data, and k represents the number of data collected by the sensor in one day; Input the feature matrix into the prediction model for initial training, calculate the mean square error MSE based on the predicted value and the measured value of the 31st day output by the prediction model, and retain the prediction model parameters when the mean square error MSE is the smallest as the optimal parameters. The expression of the mean square error MSE is as follows: Among them, x i is the measured value, is the predicted value output by the prediction model, and n is the number of data; Adopt the Pauta criterion to classify the predicted value of the 31st day. When the classification result shows normal, directly retrain the model with the measured values from the 2nd day to the 31st day and then predict the data of the 32nd day; when the classification result shows abnormal, retrain the model with the measured values from the 2nd day to the 30th day and the predicted value of the 31st day and then predict the data of the 32nd day.

5. A computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the method described in any one of claims 1-4.

6. A computer storage medium, characterized in that, Instructions are stored in the computer storage medium, and when the instructions are executed on the computer, the computer is caused to execute the method described in any one of claims 1-4.

Images