A big data driven intelligent pipe network scheduling system and method based on demand

Abstract

The application discloses a big data driven intelligent pipe network scheduling system and method, which comprises an access gateway module, a data standardization module, a database, an AI engine module and a Portal module. The AI engine has data prediction and analysis capabilities through a neural network deep learning algorithm. First, the AI engine performs model training by calling historical pipe network data. Then, the AI engine predicts pipe network operation data in a future period of time by calling real-time data at the current time. According to the prediction result and in combination with the operation condition of a water pump, a quantitative scheduling scheme close to an optimal solution is provided for scheduling of the water pump of a water plant and a booster station. Under the premise of keeping the water level of a regulating pool safe and reasonable and guaranteeing water supply stability, the booster water pump is reasonably scheduled, the service life of the pump is prolonged, and energy consumption is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of pipe network scheduling, in particular to a big data driven intelligent pipe network scheduling system and method. BACKGROUND

[0002] The water supply of a water plant is through a pipe network composed of various levels of pipes laid in a city, connected to users for water supply. A water company needs to use multi-stage pressurized water pumps to increase the water pressure into the pipe network, and also needs to ensure the pressure balance of the pipe network. The pipe network pressure cannot be too high to cause pipe explosion, and the pipe network pressure cannot be too low to cause insufficient water supply pressure at the lowest point of the user. It is also necessary to adjust the pipe network water supply according to the amount of water used by the user, and to perform emergency treatment by maintaining the liquid level of the regulating pool. Therefore, balancing the water supply pressure, water supply and water pump energy consumption is a difficult point of the water supply pipe network. At present, the water supply methods used by various water plants are as follows:

[0003] The first kind: manual scheduling is mostly used, which requires water supply scheduling personnel to have certain experience, to schedule the pipe network according to the pressure monitoring points on the pipe network and their understanding of the user's water usage habits. The problem of this method is that the scheduling method is extensive, causing the pipe network pressure to be too high, the pipe network to have the risk of pipe explosion, the regulating pool liquid level to be kept at a high or low level, and the water supply pump house water pump scheduling to be not optimal, resulting in high energy consumption.

[0004] The second kind: pipe network hydraulic model scheduling: many water plants realize the problems of manual scheduling and hope to use hydraulic model method for optimization. The hydraulic model method is mainly based on mechanism model, which reacts the pipe network state on the computer through pipe network survey to simulate the operation of the pipe network. The advantages are that the modeling conforms to the actual situation, the operation is accurate and precise, and various conditions of the pipe network can be simulated in real time. The disadvantages are that the implementation is difficult, the construction period is long, the survey cost is high, the maintenance is difficult, errors are inevitable in the survey process, the model accuracy decreases if the pipe construction or changes are not maintained, the software maintenance requires high skills of personnel, and professional personnel are needed for maintenance and other work. SUMMARY

[0005] In order to solve the above technical problems, the present application provides the following technical solutions:

[0006] A big data-driven intelligent pipeline network scheduling system predicts water demand using neural network algorithms, and rationally schedules water pumps based on demand. The system includes an access gateway module, a data standardization module, a database, an AI engine module, and a portal module. The access gateway module collects real-time data from IoT smart devices. These devices connect to the access gateway module via a network and exchange uplink and downlink messages using a transmission protocol. The data standardization module includes real-time standardization and timed standardization. Real-time standardization processes the collected data, providing the standardized data for prediction by the AI ​​engine module and storing it in the database. Standardization removes outlier data and verifies reported data based on device type and data characteristics. Timed standardization reads standardized data from various devices stored in the database, unifying historical data from devices with different collection frequencies into frequency D characteristic data, which is then stored in the database and used as training data for various model training modules of the AI ​​engine. The database stores IoT smart device data and weather data collected by the access gateway module, standardized data processed by the data standardization module, prediction data from the AI ​​engine module, water supply area data, and regulating tank level range data. The AI ​​engine module uses neural network algorithms to train models by calling historical data from the pipeline network and to predict and analyze real-time data from the pipeline network, providing scheduling solutions. It includes a water pump supply and intelligent scheduling module, a water demand prediction module, and a regulating tank level prediction module. The water pump supply and intelligent scheduling module is used to intelligently schedule water pumps in each water supply area and predict water supply volume. The water demand prediction module predicts water consumption for each water supply area by using a time series neural network algorithm to train a model using historical data for that area, and then uses the trained model and real-time data to predict water demand for a future period. The regulating tank level prediction module reads standardized historical data, uses a multi-layer time series neural network algorithm to train a model, and then uses the trained model and real-time data to predict the regulating tank level for a future period. The Portal module displays real-time, historical, and analytical data from IoT smart devices in each process.

[0007] As a preferred technical solution of the present invention, the uplink message refers to the data indicators collected from IoT smart devices and the weather data obtained from the Internet; the downlink message refers to the message generated by the intelligent scheduling module, which is sent to the water pump through the access gateway module to remotely control the start and stop of the water pump and the frequency setting.

[0008] The method for real-time standardization is as follows: using H-day historical data to calculate the standard deviation, judging that the absolute value of the adjacent data difference exceeding M times of the standard deviation is abnormally increased or decreased, the collected data less than 0 is abnormal data, deleting the abnormal data, and storing the normal data in the database intermediate table;

[0009] The algorithm for unifying the frequency D in the timing standardization is as follows:

[0010] The flow is the cumulative flow in each D time;

[0011] The pressure is the average pressure in each D time;

[0012] The pump frequency is the average frequency in each D time;

[0013] The liquid level is the average liquid level in each D time;

[0014] The storage capacity of the regulating pool is the liquid level difference X the bottom area of the regulating pool in D time;

[0015] The weather is the average temperature in each D time;

[0016] The liquid level difference is the average liquid level in each D time minus the average liquid level in (D-1) time;

[0017] The water demand is the flow minus the storage capacity of the regulating pool

[0018] The weather data is real-time temperature data obtained from the Internet;

[0019] As a preferred technical solution of the present application, in the real-time standardization method, H days are 30 days, M times are 3 times, and D is 1 hour; the real-time standardized data includes flow data, pressure data, frequency data, and liquid level data;

[0020] As a preferred technical solution of the present application, the water pump water supply and intelligent scheduling module includes a water pump water supply model training module, a water pump water supply data prediction module, and a water pump intelligent scheduling module;

[0021] The water pump water supply model training module is used to use the standardized data as model training data, use a feedforward neural network to train the model, save the trained model in a specified model file directory, call the water pump water supply data prediction module, set the update time, update the used data, and execute the model training to generate a new model at the same time;

[0022] The water pump water supply data prediction module reads the future N-hour water demand data predicted by the water demand prediction module and the current water pump frequency real-time data in the database, and predicts the future N-hour water output;

[0023] The water pump intelligent scheduling module is used for intelligently scheduling the water pump according to the water supply area, intelligently calculating according to the prediction data of the regulating tank and the prediction data of the water demand, and the water pump intelligent scheduling module is set to be executed once every W hours in the program;

[0024] The water demand prediction module comprises a water demand prediction model training module and a water demand data prediction module.

[0025] The water demand prediction model training module reads historical data of the water demand, the outlet water pressure, the regulating tank liquid level difference, the weather, the day of the week and the like after being processed by the standardization module in the database, sorts the data according to time to form a time sequence, uses a multilayer time sequence neural network algorithm to perform model training, saves the trained model in a specified model file directory for calling by the data prediction module, sets an update time to update the used data and simultaneously perform model training to generate a new model;

[0026] The water demand data prediction module uses real-time data of the water demand, the regulating tank liquid level difference, the weather, the day of the week and the like after being processed by the data standardization module, calls the trained model of the water demand prediction model training module to predict the water demand in a future period of time, and stores the prediction result in the database.

[0027] The regulating tank liquid level prediction module comprises a regulating tank liquid level prediction model training module and a regulating tank liquid level data prediction module.

[0028] The regulating tank liquid level prediction model training module reads historical data of the water plant outlet water flow, the water plant outlet water pressure, the regulating tank liquid level, the weather, the day of the week and the like after being processed by the standardization module in the database, sorts the data according to time to form a time sequence, uses a multilayer time sequence neural network algorithm to perform model training, saves the trained model in a specified model file directory for calling by the data prediction module, sets an update time to update the used data and simultaneously perform model training to generate a new model.

[0029] The regulating tank liquid level data prediction module uses real-time data of the water plant outlet water flow, the water plant outlet water pressure, the regulating tank liquid level, the weather, the day of the week and the like after being processed by the data standardization module, calls the trained model of the regulating tank liquid level prediction model training module to predict the regulating tank liquid level in a future period of time, and stores the prediction result in the database.

[0030] As a preferred technical solution of the present application, the number of models trained by the water pump water supply model training module is determined according to the water pump type:

[0031] The first case: both are variable frequency pumps, read the number L of variable frequency pumps in the water pump information set in the current water supply area management module, read the pump frequency, water demand data and pump unit efficiency processed by the database standardization module as input indicators for model training, establish L models, and establish models according to the number of pumps in the combined pump unit combination;

[0032] The second case: both are fixed frequency pumps, read the number of fixed frequency pumps under the current water supply area, read the historical data processed by the database standardization module, the number of pump operation, water demand and pump unit efficiency as input indicators, and the water plant output as output;

[0033] The third case: both are variable frequency pumps, read the number L of variable frequency pumps in the water pump information set in the current water supply area management module, read the pump frequency, water demand data and pump unit efficiency processed by the database standardization module as input indicators for model training, establish L models, and establish models according to the number of pumps in the combined pump unit combination, the number of fixed frequency pumps does not affect the number of models, and the number of fixed frequency pumps is used as an input indicator;

[0034] The water pump intelligent scheduling module includes a regulating pool liquid level prediction algorithm and a water pump scheduling algorithm;

[0035] The regulating pool liquid level prediction algorithm calculates the liquid level in the future N hours according to the future N hours of water supply predicted by the water supply data prediction module and the future N hours of water demand predicted by the water demand data prediction module, and the algorithm is as follows:

[0036] The regulating pool predicted water storage capacity = the sum of the predicted future N hours of water supply - the sum of the future N hours of water demand,

[0037] The regulating pool predicted liquid level difference = the regulating pool predicted water storage capacity / the regulating pool bottom area,

[0038] The future N hours liquid level = the current regulating pool real-time liquid level + the regulating pool predicted liquid level difference

[0039] The water pump scheduling algorithm adjusts the frequency of the variable frequency pump or the number of fixed frequency pumps by setting the whole point time i in the future N hours, and re-calls the water pump water supply data prediction module and the water demand data prediction module to perform water supply prediction and water demand prediction, respectively. This algorithm is used to predict the change of the regulating pool liquid level in the future N hours according to the scheduling of the water pump, and the regulating pool liquid level reaches the safe liquid level range before i reaches N, while generating device operation instructions and sending them to the device;

[0040] The algorithm is as follows:

[0041] First step: determine whether the regulating pool liquid level is in a reasonable range according to the current pump operation;

[0042] The water demand predicted by the current running pump operation condition and water demand data prediction module is taken as input, the water supply amount predicted by the water pump water supply data prediction module is called, the regulating pool liquid level in the future N hours is calculated by calling the regulating pool liquid level prediction algorithm, and whether it is within a reasonable range is judged;

[0043] If it is not within a reasonable range, the second step is performed,

[0044] If it is within a reasonable range, the scheduling information is returned;

[0045] Second step: screening the pump to be operated

[0046] When the number of currently running pumps is 1, the current pump is selected as the pump to be operated;

[0047] When the number of currently running pumps is greater than 1: the efficiency priority algorithm is called for operation to screen the pump to be adjusted;

[0048] Third step: operating according to the selected pump type

[0049] First, the selected pump is a variable frequency pump: the frequency of the selected pump is adjusted according to a step P,

[0050] When it exceeds the range, the pump frequency is lowered, and when it is lower than the range, the pump frequency is increased;

[0051] The frequency of the selected pump is adjusted, and the pump frequency is increased or decreased by P to obtain the adjusted pump operation condition. Set i = 1, i <= N, from the ith hour, the water demand predicted by the current pump operation condition and water demand data prediction module is taken as input before i hours, and the water demand predicted by the current pump operation condition and water demand data prediction module is taken as input after i hours. The water pump water supply data prediction module and the water pump intelligent scheduling module are called N times, the regulating pool liquid level prediction algorithm is used to calculate the regulating pool liquid level in the future N hours, and whether it is within a reasonable range is judged. If it is still not within a reasonable range, the frequency of the selected pump is increased or decreased by P, and then whether it is within a reasonable range is judged. Each time P is increased or decreased, until it is judged to be within a reasonable range, and the scheduling information is returned;

[0052] If the following two situations occur during the adjustment process, they are handled respectively:

[0053] The selected pump frequency is adjusted to or below the minimum value, and the selected pump needs to be stopped. Return to the first step of operation;

[0054] The selected pump frequency is adjusted to or above the maximum value, and the second step of operation is returned to increase one pump. If the number of currently running pumps is 0, the scheduling information is returned. If all pumps are in a running state and the frequency of each pump is adjusted to or exceeds the maximum value, the scheduling information is returned;

[0055] Second, the selected pump is a fixed frequency pump: adjust the number of running pumps, stop the selected pump when it exceeds the range, and start the selected pump when it is below the range;

[0056] The number of pumps to be stopped or started is input, i = 1, i <= N, from the ith hour, the predicted water demand of the current pump operation and water demand data prediction module before i hours is input, and the predicted water demand of the adjusted pump operation and water demand data prediction module after i hours is input, the water pump supply data prediction module and the water pump intelligent scheduling module are called N times, the adjusted pool liquid level prediction algorithm is used to calculate the future N hours of the adjusted pool liquid level, and it is judged whether it is within a reasonable range; If it is within a reasonable range, return the scheduling information; If it is still not within a reasonable range, return to the second step; If the number of running pumps is 0 or all, return the scheduling information;

[0057] Fourth, generate device operation instructions and send them to the device

[0058] According to the returned scheduling information, generate a downlink control instruction, remotely control through the access gateway module, send the downlink control instruction to the pump that needs to be remotely controlled at the current actual time plus i hours, and perform frequency setting, pump starting / stopping and other operations;

[0059] The downlink control instruction is a message sent to the water pump, which can be recognized and executed by the water pump, and the specific format is provided by the water pump manufacturer or the communication module manufacturer on the water pump;

[0060] The efficiency priority algorithm: according to the historical data of the water pump, the water pump unit efficiency algorithm is calculated, and the combination mode with high efficiency is preferentially selected; When there are E running pumps and F non-running pumps, the following cases are considered:

[0061] When the pump needs to be adjusted: any X water pumps in the running pumps are combined, the water pump unit efficiency algorithm is called to sort the single pump efficiency of the X water pumps, and the pump not adjusted to the highest or lowest is selected in order;

[0062] When the pump needs to be stopped: any X water pumps in the running pumps are combined, the water pump unit efficiency algorithm is called to calculate the unit efficiency, and the combination with the highest unit efficiency is retained. The remaining 1 water pump is the selected pump, and X = E-1;

[0063] When the pump needs to be increased: all running pumps and non-running pumps are combined, i.e. Z pumps are combined, the unit efficiency is calculated, and the combination with the highest unit efficiency is retained, where the non-running pump is the selected pump, and Z = E+1;

[0064] The water pump unit efficiency algorithm:

[0065] Output power = (instantaneous flow x 1000 / 3600) x (pressure x 100 - regulating pool level)

[0066] Single pump input power = voltage x current x 1.732 x 0.95

[0067] Pump unit output power sum = Σ single pump input power

[0068] Water pump unit efficiency = output power / pump unit output power sum

[0069] As a preferred technical solution of the present application, the Portal module includes a water intake pump house module, a clear water pool module, a water delivery pump house module, a regulating pool module, a today's water supply situation module, a source water level module, a pipe network data module, a water quality monitoring module, a data analysis module, a water supply area management module, and a regulating pool level range setting module.

[0070] The water intake pump house module is used to display real-time data and historical data of the water intake pump operation, and the data indexes include operation condition (running / shutdown), running / shutdown time, rated flow, and frequency. The historical data of each index in the selected time period can be queried from the database by selecting the time range, and the data is read from the database.

[0071] The clear water pool module is used to display real-time data and historical data of the liquid level instrument, and the data indexes include liquid level. The historical data of the liquid level can be queried from the database according to time, and displayed on the page in the form of a curve. The data is read from the database.

[0072] The water delivery pump house module is used to display real-time data and historical data of the water delivery pump operation, and the data indexes include operation condition (running / shutdown), running / shutdown time, rated flow, and frequency. The historical data of each index in the selected time period can be queried from the database by selecting the time range, and the data is read from the database.

[0073] The today's water supply situation module includes today's water demand, today's water supply, and today's water storage, and is used to display and calculate today's water demand, today's water supply, and today's water storage. The data is read from the database, and the calculation method is as follows:

[0074] Today's water demand = today's water supply - today's water storage

[0075] Today's water supply refers to the cumulative flow of water plant outflow from 0 o'clock to the current time

[0076] Today's water storage refers to the water demand of the regulating pool from 0 o'clock to the current time.

[0077] The source water module is used to display real-time data and historical data of a source water level instrument on a page, and the data indexes include a source water level, and historical conditions of the source water level can be queried from a database according to time, and the historical conditions are displayed in a form of a curve on the page, and data is read from the database;

[0078] The pipe network data module reads pressure data and position data of key nodes of a pipe network collected by the access gateway module from a database, the position data is longitude and latitude information of an installation position of a pressure device, and is displayed on a page. Historical pressure conditions of each key node of the pipe network can be queried from the database according to time, and are displayed on the page. Historical maximum pressure and minimum pressure thresholds are found out by analyzing historical data of the key nodes of the pipe network, and abnormal condition alarm analysis is performed on real-time data, such as high pressure and low pressure alarms. When an alarm of a pipe network node occurs, the pipe network node is displayed in red on the page, and a high pressure or low pressure alarm is prompted;

[0079] The water quality monitoring module is used to display real-time water quality data and historical data trends of each process section of the pipe network, including a water quality module, a turbidity trend module and a residual chlorine trend module.

[0080] The water quality module reads real-time data of turbidity, PH value, residual chlorine and other water quality related indexes of main process sections from a database, and displays the real-time data on a page. Process sections displayed on the page include raw water quality, sedimentation tank water quality, filtered water quality and finished water quality.

[0081] The turbidity trend module reads historical turbidity data of process sections such as a sedimentation tank, filtered water and finished water from a database, and displays turbidity trends of the process sections on a page.

[0082] The residual chlorine module reads historical residual chlorine data of process sections such as filtered water and finished water from a database, and displays residual chlorine trends of the process sections on a page.

[0083] The data analysis module is used to analyze data prediction accuracy of water supply prediction, water demand prediction, regulating pool liquid level prediction and the like, including a water supply prediction analysis module, a water demand prediction analysis module, a clear water pool prediction analysis module and a regulating pool prediction analysis module.

[0084] The water supply prediction analysis module reads a water supply area from a database, reads water supply prediction data and standardized historical data from the database according to the water supply area, compares curves by chart display, and views fitting conditions of the two curves. Prediction data of N hours in the future and prediction accuracy of each hour are displayed.

[0085] The water demand prediction analysis module reads the water supply area from the database, reads the water demand prediction data and the standardized historical data from the database according to the water supply area, displays the comparison curve through the chart, and views the fitting condition; the prediction data of the future N hours and the prediction accuracy of each hour are displayed;

[0086] The regulating pool liquid level prediction analysis module reads the water supply area from the database, reads the regulating pool liquid level prediction data and the standardized historical data from the database according to the water supply area, displays the comparison curve through the chart, and views the fitting condition; the prediction data of the future N hours and the prediction accuracy of each hour are displayed;

[0087] The prediction accuracy algorithm is:

[0088] Prediction accuracy = Σ prediction accuracy of each time point / number of time points

[0089] The water supply area management module is used for creating a plurality of different water supply areas according to the actual pipe network water supply condition, the area name cannot be repeated, and when the water supply area is created, the information under the area is set, including:

[0090] Water supply plant information, including identification and name

[0091] Regulating pool information, including regulating pool identification and name, regulating pool safety height, regulating pool bottom area and the like

[0092] Pump information, including pump identification, name, whether frequency conversion, maximum frequency and minimum frequency

[0093] The regulating pool liquid level range setting module is used for setting the regulating pool liquid level safety range in different seasons, including: the start time and the end time of each season, a plurality of time periods are set in each season, and the lowest value and the highest value of the liquid level corresponding to each time period; according to the characteristics of the seasons, 1 day is divided into several time periods such as water supply peak and valley, the start time and the end time of each time period are input in the time format HH:MM, a plurality of time periods can be added, the time periods cannot be crossed and cover 0-24 hours, a reasonable safety liquid level range value is set for each time period, the lowest value of the liquid level is greater than 0, the highest value of the liquid level is less than the safety height of the regulating pool, the range cannot be input, and is stored in the database.

[0094] As a preferred technical scheme of the present application, N is 6; the update time is set to 0 o'clock every day in the model training of the AI engine module; W is 1; and P is 3.

[0095] A big data driven intelligent pipe network scheduling method for demand-based supply uses a neural network algorithm to intelligently control the intelligent scheduling of pipe network water supply, and the steps are as follows:

[0096] Establishing a water demand prediction LSTM neural network model

[0097] Collect the cumulative flow of the effluent of the water supply plant in the water supply area, the liquid level difference of the regulating pool, the weather, the day of the week and other data as the training input vector, determine the safety range of the regulating pool on the pipe network according to experience, establish the training set of the water demand prediction, train the water demand prediction LSTM neural network by using the water demand training set to obtain the water demand prediction model, and predict the water demand by using the real-time data vector through the water demand prediction model;

[0098] Establish a water pump supply BP neural network model

[0099] Collect the running condition and power of the frequency conversion pump, the water demand, the number of fixed frequency pumps and other data in the water supply area as the input vector, collect the effluent data of the water plant as the output vector, establish the training set of the water pump supply prediction including the input vector and the output vector, and train the water pump supply BP neural network model by using the water pump supply prediction training set to obtain the water supply prediction model;

[0100] Intelligent scheduling method

[0101] The water demand is predicted by using the real-time data of the water demand through the trained water demand prediction LSTM model, the water supply is predicted by using the real-time data vector of the water supply through the trained water supply prediction BP model, the optimal scheduling scheme is obtained through the intelligent scheduling algorithm, and the scheme is regulated and controlled;

[0102] As a preferred technical scheme of the present application, the LSTM neural network model:

[0103] The first layer LSTM uses 128 cell units, solves overfitting by using a dropout layer, and sets the dropout probability parameter to 0.1.

[0104] The second layer LSTM uses 64 cell units, uses Sigmoid as the activation function, solves overfitting by using a dropout layer, and sets the dropout probability parameter to 0.5.

[0105] The third layer LSTM uses 32 cell units, uses Sigmoid as the activation function, solves overfitting by using a dropout layer, and sets the dropout probability parameter to 0.5.

[0106] As a preferred technical scheme of the present application, the BP neural network model, the first layer Dense layer uses 64 cell units, and uses relu as the activation function; the second layer Dense layer uses 64 cell units. BRIEF DESCRIPTION OF DRAWINGS

[0107] The accompanying drawings are used to provide further understanding of the present application, and constitute a part of the specification, together with embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0108] Figure 1 is a system structure diagram of the present application;

[0109] Figure 2 is a process diagram of establishing a water demand prediction LSTM neural network model of the present application;

[0110] Figure 3 is a process diagram of establishing a water pump water supply BP neural network model of the present application;

[0111] Figure 4 is a process diagram of an intelligent scheduling method of the present application;

[0112] Figure 5 is a schematic diagram of an LSTM neural network algorithm of the present application;

[0113] Figure 6 is a schematic diagram of a BP neural network algorithm. DETAILED DESCRIPTION

[0114] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application. EMBODIMENT

[0115] As shown in Figures 1-6 , the present application is an intelligent pipe network scheduling system driven by big data and based on demand, which comprises an access gateway module, a data standardization module, a database, an AI engine module, and a Portal module. The AI engine has data prediction and analysis capabilities through a neural network deep learning algorithm, can call historical pipe network data for model training, call real-time data at the current time of the pipe network for data prediction of the pipe network operation in the future, and combine the operation of the water pump to provide a quantitative and near-optimal scheduling scheme for the scheduling of the water pump of the water plant and the booster station, reasonably start and stop the booster water pump under the premise of maintaining the water level of the regulating pool and ensuring water supply, prolong the service life of the pump, and save energy and reduce consumption.

[0116] The access gateway module is used to collect real-time data of Internet of Things intelligent devices installed in various process procedures of the water plant, such as source water, clean water pool, secondary pump station, and pipe network. These Internet of Things intelligent devices are linked to the access gateway module through a network, and perform uplink message and downlink message interaction with the access gateway module through transmission protocols such as OPC, TCP / UDP, etc.

[0117] Specifically, the uplink message refers to the data index collected from the Internet of Things intelligent device and weather data obtained from the Internet; the data index includes cumulative flow, water pressure, pipe network key node pressure, regulating pool liquid level difference, intelligent water pump operation parameter, running state, frequency and other data of the water supply plant effluent; the data collected from the device is directly stored into the database according to the device type, and is standardized by the data standardization module.

[0118] Specifically, the downlink message refers to the message generated by the intelligent scheduling module and sent to the water pump through the access gateway module, which is used to remotely control the start and stop and frequency setting of the water pump.

[0119] The data standardization module includes real-time standardization and timing standardization.

[0120] The real-time standardization is used for standardizing the data collected by the access gateway module, and the standardized data is used for prediction by the AI engine module and is also stored in the database. The standardization process is to remove abnormal data, and to check the reported data according to the device type and data characteristics.

[0121] The real-time standardization method of each type of data is as follows:

[0122] The standard deviation is calculated using H-day historical data, and the absolute value of the difference between adjacent data exceeding M times the standard deviation is considered as abnormal increase or decrease. The collected data less than 0 is considered as abnormal data, and the abnormal data is deleted. The normal data is stored in the database intermediate table.

[0123] Specifically, H days are 30 days (starting from the current date, counting back 30 days as sample data, and calculating the standard deviation of the previous 30 days once a day at 0 o'clock), and 3 times M is the optimal scheme for judging abnormality.

[0124] Specifically, the data subjected to real-time standardization includes flow data, pressure data, frequency data and liquid level data.

[0125] Timing standardization is to read the standardized data of each device stored in the database, unify the historical data of each device with different collection frequencies into frequency D characteristic data, and store it into the database as training data for each model training module of the AI engine. The algorithm for unifying each device historical data into frequency D is as follows:

[0126] The flow is the cumulative flow per D time;

[0127] The pressure is the average pressure per D time;

[0128] The water pump frequency is the average frequency per D time;

[0129] The liquid level is the average liquid level per D time;

[0130] The regulating pool water storage amount = liquid level difference per D time X regulating pool bottom area;

[0131] The weather is the average temperature per D time;

[0132] The liquid level difference = average liquid level per D time - average liquid level per (D-1) time;

[0133] The water demand = flow - regulating pool water storage amount;

[0134] Specifically, D is 1 hour which is the optimal choice;

[0135] The database is used to store the Internet of Things intelligent device data and weather data collected by the access gateway module, the standardized data processed by the data standardization module, the prediction data of the AI engine module, the water supply area data, the regulating pool liquid level range data, etc.

[0136] The weather data is real-time temperature data obtained from the Internet;

[0137] The following is an explanation of real-time data and historical data:

[0138] The real-time data refers to the latest data in the Internet of Things intelligent device data and weather data stored in the database, and the latest data in the standardized data processed by the standardization module.

[0139] The historical data refers to the non-latest data in the Internet of Things intelligent device data and weather data stored in the database, and the non-latest data in the standardized data processed by the standardization module.

[0140] The AI engine module is the core module of the water supply network intelligent scheduling system, which is used to predict water demand, regulating pool liquid level and perform water pump intelligent scheduling according to the water supply area, so as to achieve the effect of energy saving and consumption reduction according to the demand. The AI engine module includes a water pump water supply and intelligent scheduling module, a water demand prediction module, and a regulating pool liquid level prediction module;

[0141] The water pump water supply and intelligent scheduling module is used to intelligently schedule and predict the water supply of the water pump under each water supply area, including a water pump water supply model training module, a water pump water supply data prediction module, and a water pump intelligent scheduling module;

[0142] The water pump water supply model training module is configured to use the standardized data as model training data, use a feedforward neural network to perform model training, save the trained model in a specified model file directory, and call the water pump water supply data prediction module. Model training needs to maintain the optimization of the model, and the data used needs to be updated. The update time is set in the program, and a new model is generated once the model training is performed at the update time.

[0143] Specifically, the update time is set to 0 o'clock every day as the optimal solution, or it can be set according to actual needs.

[0144] The number of models trained by the water pump water supply model training module is determined according to the type of water pump:

[0145] The first case: all variable frequency pumps, read the number L of variable frequency pumps in the water pump information set in the current water supply area management module, read the pump frequency, water demand data and pump unit efficiency processed by the standardization module in the database as input indicators for model training, and establish L models according to the number of pumps in the combined pump unit combination.

[0146] For example, there are 3 variable frequency pumps, and the models are established as follows:

[0147] Model one: one pump running

[0148] Pump one frequency, water demand as input indicators, pump unit efficiency, water plant output as output.

[0149] Model two: two pumps running

[0150] Pump one frequency, pump two frequency, pump unit efficiency, water demand as input indicators, water plant output as output.

[0151] Model three: three pumps running

[0152] Pump one frequency, pump two frequency, pump three frequency, pump unit efficiency, water demand as input indicators, water plant output as output.

[0153] The second case: all fixed frequency pumps, read the number of fixed frequency pumps under the current water supply area, read the historical data processed by the standardization module in the database, pump running number, water demand, pump unit efficiency as input indicators, and water plant output as output.

[0154] The third case: all variable frequency pumps, read the number L of variable frequency pumps in the water pump information set in the current water supply area management module, read the pump frequency, water demand data, and pump unit efficiency processed by the standardization module in the database as input indicators for model training, establish L models according to the number of pumps in the combined pump unit combination, and the number of fixed frequency pumps does not affect the number of models. The number of fixed frequency pumps is used as an input indicator.

[0155] For example: 2 variable frequency pumps and 2 fixed frequency pumps are combined to establish a model as follows:

[0156] Model one: 1 variable frequency pump runs

[0157] Pump one frequency, water demand, pump unit efficiency, and fixed frequency pump number are input indicators, and water plant output is output.

[0158] Model two: 2 variable frequency pumps run

[0159] Pump one frequency, pump two frequency, water demand, pump unit efficiency, and fixed frequency pump number are input indicators, and water plant output is output.

[0160] BP is the abbreviation of feedforward neural network. BP neural network is a multilayer feedforward network trained by error backpropagation, which is abbreviated as error backpropagation. Its basic idea is gradient descent method, and gradient search technology is used to minimize the mean square error of the actual output value and the expected output value of the network. Based on BP neural network, through normalization, BP, and inverse normalization algorithms, a neural network is formed;

[0161] The feature data is composed of a record, multiple column inputs, and single column outputs.

[0162] Normalization operation is performed, that is, the feature data is mapped to (0, 1) for processing, which is faster and improves processing performance.

[0163] The first layer Dense (full connection layer) is the input layer, which uses D1 (64 is the optimal solution after testing) cell units, and the activation function uses relu (rectified linear unit, which is a commonly used activation function in neural networks).

[0164] The second layer Dense (full connection layer) is the input layer, which uses D2 (64 is the optimal solution after testing) cell units, and the activation function uses relu.

[0165] The third layer Dense (full connection layer) is the output layer, which has only one cell unit, and the output column is 1.

[0166] Inverse normalization operation is performed, that is, the result data between (0, 1) obtained by prediction is restored to feature data through the inverse normalization function.

[0167] The prediction result is output.

[0168] The water pump water supply data prediction module reads the future N-hour water demand data predicted by the water demand prediction module and the current water pump frequency real-time data in the database, calls the model trained by the water pump water supply model training module, predicts the future N-hour water output, and saves the prediction result in the database:

[0169] Specifically, N is 6 is the optimal scheme

[0170] The water pump intelligent scheduling module is used for intelligent scheduling of water pumps according to water supply areas, and intelligent calculation is performed according to the regulation pool prediction data and the water demand prediction data, including a regulation pool liquid level prediction algorithm and a water pump scheduling algorithm. The water pump intelligent scheduling module is set to execute once every W hours in the program, and 1 hour is the optimal scheme.

[0171] The regulation pool liquid level prediction algorithm calculates the liquid level in the future N hours according to the future N-hour water supply predicted by the water supply data prediction module and the future N-hour water demand predicted by the water demand prediction module. The algorithm is as follows:

[0172] The regulation pool predicted water storage capacity = the sum of the predicted future N-hour water volume - the sum of the future N-hour water demand,

[0173] The regulation pool predicted liquid level difference = the regulation pool predicted water storage capacity / the regulation pool bottom area,

[0174] The future N-hour liquid level = the current regulation pool real-time liquid level + the regulation pool predicted liquid level difference

[0175] The water pump scheduling algorithm adjusts the frequency of the variable frequency water pump or the number of fixed frequency pumps by setting the integral time i in the future N hours, re-calls the water pump water supply data prediction module and the water demand data prediction module to perform water supply prediction and water demand prediction respectively. This algorithm is used to predict the regulation pool liquid level change in the future N hours according to the scheduling of the water pump, and the regulation pool liquid level reaches the safe liquid level range before i reaches N, while generating device operation instructions and sending them to the device.

[0176] The water pump scheduling algorithm:

[0177] Step 1: Determine whether the regulation pool liquid level is in a reasonable range according to the current pump operation

[0178] Use the current pump operation, the water demand predicted by the water demand data prediction module as input, call the water supply predicted by the water pump water supply data prediction module, call the regulation pool liquid level prediction algorithm to calculate the regulation pool liquid level in the future N hours, and determine whether it is in a reasonable range;

[0179] Not in a reasonable range, go to step 2,

[0180] In a reasonable range, return the scheduling information;

[0181] Second step: screening the pump to be operated

[0182] If the number of currently running pumps is 1, the current pump is selected as the pump to be operated;

[0183] If the number of currently running pumps is greater than 1, the efficiency priority algorithm is called to operate and screen the pump to be adjusted;

[0184] Third step: operating according to the selected pump type

[0185] First, the screened pump is a variable frequency pump: the frequency of the selected pump is adjusted according to a step P,

[0186] When it exceeds the range, the pump frequency is lowered, and when it is lower than the range, the pump frequency is increased;

[0187] Adjust the frequency of the screened pump, increase or decrease the pump frequency by P to obtain the adjusted pump operation, set i = 1, i <= N, from the ith hour, use the current pump operation, the water demand predicted by the water demand data prediction module as input before i hours, and use the adjusted pump operation, the water demand predicted by the water demand data prediction module as input after i hours, call the water pump supply data prediction module and the water pump intelligent scheduling module N times, adjust the pool liquid level prediction algorithm to calculate the pool liquid level in the next N hours, and judge whether it is within a reasonable range; if it is still not within a reasonable range, continue to increase or decrease the frequency of the selected pump by P, and then judge whether it is within a reasonable range, increase or decrease P each time until it is within a reasonable range, and return the scheduling information;

[0188] If the following two situations occur during the adjustment process, they are handled respectively:

[0189] The selected pump frequency is adjusted to or below the minimum value, the selected pump needs to be stopped, and the first step is returned to operation;

[0190] The selected pump frequency is adjusted to or above the maximum value, and the second step is returned to operation to increase one pump; if the number of currently running pumps is 0, the scheduling information is returned; if all pumps are in running state and the frequency of each pump is adjusted to or exceeds the maximum value, the scheduling information is returned;

[0191] Second, the screened pump is a fixed frequency pump: the number of running pumps is adjusted, when it exceeds the range, the selected pump is stopped, and when it is lower than the range, the selected pump is started;

[0192] The number of pumps to be stopped or started is taken as input, i = 1, i <= N, from the i hour, the water demand predicted by the pump operation condition and water demand data prediction module before i hours is taken as input, and the water demand predicted by the pump operation condition and water demand data prediction module after adjustment is taken as input, the water pump supply data prediction module and the water pump intelligent scheduling module are called N times, the pool liquid level prediction algorithm is used to calculate the pool liquid level in the next N hours, and whether it is within a reasonable range is judged; if it is within a reasonable range, return the scheduling information; if it is still not within a reasonable range, return to the second step; if the number of running pumps is 0 or all running, return the scheduling information;

[0193] Fourthly, device operation instructions are generated and sent to the device

[0194] According to the returned scheduling information, downlink control instructions are generated, which are sent to the pump to be remotely controlled through the access gateway module at the current actual time plus i hours, and frequency setting, pump starting / stopping and other operations are performed.

[0195] Specifically, the minimum frequency and the maximum frequency of the variable frequency water pump are set in the water supply area management module.

[0196] The downlink control instruction is a message sent to the water pump, which can be recognized and executed by the water pump, and the specific format is provided by the water pump manufacturer or the communication module manufacturer on the water pump

[0197] Specifically, according to the performance curve set provided by the variable frequency pump manufacturer, it is tested that the step P of 3 is the optimal scheme.

[0198] The efficiency priority algorithm: according to the historical data of the water pump, the water pump unit efficiency algorithm is calculated, and the combination mode with high efficiency is preferentially selected; when there are E running pumps and F non-running pumps, the following cases are divided:

[0199] When the pump to be adjusted is selected: any X water pumps in the running pumps are combined, the water pump unit efficiency algorithm is called to sort the single pump efficiency of the X water pumps, and the pump not running at the highest or lowest is selected in order;

[0200] When the pump to be stopped is selected: any X water pumps in the running pumps are combined, the water pump unit efficiency algorithm is called to calculate the unit efficiency, the combination with the highest unit efficiency is retained, and the remaining 1 water pump is the selected pump, X = E-1;

[0201] When the pump to be added is selected: all running pumps and non-running pumps are combined, that is, Z pumps are combined, the unit efficiency is calculated, the combination with the highest unit efficiency is retained, and the non-running pump is selected as the selected pump, Z = E+1;

[0202] For example: the efficiency of pump one and pump two unit is 60%, the efficiency of pump two and pump three unit is 56%, the efficiency of pump one and pump three unit is 58%, when stopping a pump, the combination of pump one and pump two is preferred, and pump three is preferred to be stopped;

[0203] The water pump unit efficiency algorithm:

[0204] Output power = (instantaneous flow X 1000 / 3600) X (pressure X 100-adjusted pool level)

[0205] Single pump input power = voltage X current X 1.732 X 0.95

[0206] Unit pump output power sum = Σ single pump input power

[0207] Water pump unit efficiency = output power / unit pump output power sum

[0208] The water demand prediction module is used to predict the water demand of the area according to the water supply area, uses time series neural network algorithm, trains the model through the historical data of the area, and uses the trained model and real-time data to predict the water demand in the future N hours. The water demand prediction module includes a water demand prediction model training module and a water demand data prediction module.

[0209] Specifically, N is 6 is the optimal scheme, when N>6, the accuracy decreases greatly, when N<6, the predicted data time is too small, and the adjustment is too frequent;

[0210] The water demand prediction model training module reads the historical data of the standardized module processed water demand, outlet pressure, adjusted pool level difference, weather, week and other data indicators in the database, sorts the data by time to form a time series, uses a multi-layer time series neural network algorithm to train the model, and saves the trained model in a specified model file directory for calling by the data prediction module. Model training needs to keep the model optimized, and needs to be updated abnormally. According to the debugging experience, model training is executed once a day at 0 o'clock to generate a new model as the optimal.

[0211] The water demand data prediction module uses the real-time data of the water demand, the adjusted pool level difference, the weather, the week and other data indicators processed by the data standardization module, calls the trained model of the water demand prediction model training module, and predicts the water demand in the future 1 hour, 2 hours, 3 hours…N hours. The prediction results are stored in the database.

[0212] The adjusted pool level prediction module includes an adjusted pool level prediction model training module and an adjusted pool level data prediction module.

[0213] The regulating pool liquid level prediction model training module reads the historical data of the water plant effluent flow, water plant effluent pressure, regulating pool liquid level, weather, week and other data indicators processed by the standardization module in the database, sorts the data by time to form a time series, uses a multi-layer time series neural network algorithm to train the model, and saves the trained model in a specified model file directory for calling by the data prediction module. Model training needs to maintain the optimization of the model and needs to be updated abnormally. According to the debugging experience, model training is performed once a day at 0 o'clock to generate a new model as the optimal one.

[0214] The regulating pool liquid level data prediction module uses real-time data of the water plant effluent flow, water plant effluent pressure, regulating pool liquid level, weather, week and other data indicators processed by the data standardization module, calls the model trained by the regulating pool liquid level prediction model training module, predicts the regulating pool liquid level in the next 1 hour, 2 hours, 3 hours…N hours, and stores the prediction results in the database.

[0215] The Portal module is used to display real-time data, historical data, analysis data and other information of each process section, including the water intake pump house module, the clear water pool module, the water supply pump house module, the regulating pool module, the today's water supply situation module, the source water level module, the pipe network data module, the water quality monitoring module, the data analysis module, the water supply area management module, and the regulating pool liquid level range setting module.

[0216] The water intake pump house module is used to display real-time data and historical data of the operation of the water intake pump, and the data indicators include operation status (running / stop), running / stop time, rated flow, frequency, and the historical data of each indicator in the selected time period can also be selected and queried from the database.

[0217] The clear water pool module is used to display real-time data and historical data of the liquid level instrument, and the data indicators include liquid level, and the historical data of liquid level can also be queried from the database according to time and displayed on the page in the form of a curve.

[0218] The water supply pump house module is used to display real-time data and historical data of the operation of the water supply pump, and the data indicators include operation status (running / stop), running / stop time, rated flow, frequency, and the historical data of each indicator in the selected time period can also be selected and queried from the database.

[0219] The today's water supply situation module includes today's water demand, today's water supply, and today's water storage, and is used to display and calculate today's water demand, today's water supply, and today's water storage, which are read from the database, and the calculation method is as follows:

[0220] Today's water demand = today's water supply - today's water storage

[0221] Today water supply refers to the cumulative flow of water plant from 0 to the current time

[0222] Today water storage refers to the water demand of the regulation pool from 0 to the current time.

[0223] Source water module, for displaying real-time data and historical data of source water level instrument on the page, the data index includes source water level, and the historical situation of source water level can be queried from the database according to time, and displayed in the form of curve on the page.

[0224] Pipe network data module, reading the pressure data and position data of the pipe network key node collected by the gateway module from the database, the position data refers to the longitude and latitude information of the installation position of the pressure equipment, and displayed on the page. It can also be filtered according to time to query the pressure history of each pipe network key node from the database and displayed on the page. By analyzing the historical data of the pipe network key node pressure, the historical highest pressure and lowest pressure threshold value are found out, and the real-time data is analyzed for abnormal situation alarm, such as high pressure, low pressure and other alarms. When an alarm of a pipe network node occurs, the pipe network node is displayed in red on the page, and the high pressure or low pressure alarm is prompted.

[0225] Water quality monitoring module, for displaying real-time water quality data and historical data trend of each process section on the pipe network, including water quality module, turbidity trend module and residual chlorine trend module.

[0226] Water quality module, reading the real-time data of turbidity, PH value, residual chlorine and other water quality related indexes of main process section from the database and displaying on the page, the process section displayed on the page includes raw water quality, sedimentation tank water quality, filtered water quality and finished water quality.

[0227] Turbidity trend module, reading the turbidity historical data of process sections such as sedimentation tank, filtered water and finished water from the database, and displaying the turbidity trend of these process sections on the page.

[0228] Residual chlorine module, reading the residual chlorine historical data of process sections such as filtered water and finished water from the database, and displaying the residual chlorine trend of these process sections on the page.

[0229] Data analysis module, for analyzing the data prediction accuracy of water supply prediction, water demand prediction, regulation pool liquid level prediction, including water supply prediction analysis module, water demand prediction analysis module, clear water pool prediction analysis and regulation pool prediction analysis module.

[0230] The water supply prediction analysis module reads the water supply area from the database, reads the water supply prediction data and the standardized historical data from the database according to the water supply area, displays the comparison curve through the chart, and views the fitting condition of the two curves; the prediction data of the future N hours and the prediction accuracy of each hour are displayed.

[0231] The water demand prediction analysis module reads the water supply area from the database, reads the water demand prediction data and the standardized historical data from the database according to the water supply area, displays the comparison curve through the chart, and views the fitting condition; the prediction data of the future N hours and the prediction accuracy of each hour are displayed.

[0232] The regulation pool prediction analysis reads the water supply area from the database, reads the regulation pool liquid level prediction data and the standardized historical data from the database according to the water supply area, displays the comparison curve through the chart, and views the fitting condition; the prediction data of the future N hours and the prediction accuracy of each hour are displayed.

[0233] Prediction accuracy algorithm:

[0234]

[0235] Prediction accuracy = Σ prediction accuracy of each time point / number of time points

[0236] The water supply area management module is used to create multiple different water supply areas according to the actual pipe network water supply situation, and the area name cannot be repeated. When creating a water supply area, the information under the area is set, including:

[0237] Water supply plant information, including identification and name

[0238] Regulation pool information, including regulation pool identification and name, regulation pool safety height, regulation pool bottom area, etc.

[0239] Water pump information, including water pump identification, name, whether frequency conversion, maximum frequency, minimum frequency

[0240] The above information is stored in the database and used by the intelligent scheduling module of the water pump intelligent scheduling module.

[0241] The adjusting pool liquid level range setting module is used for setting the adjusting pool liquid level safety range in different seasons, and includes: the start time and the end time of each season, setting multiple time periods in each season, the lowest value and the highest value of the liquid level corresponding to each time period; according to the characteristics of the seasons, 1 day is divided into several time periods such as water supply peak and valley, the start time and the end time of each time period are input in the time format HH:MM, multiple time periods can be added, the time periods cannot be overlapped and cover 0-24 hours, a reasonable safety liquid level range value is set for each time period, the lowest value of the liquid level is greater than 0, and the highest value of the liquid level is less than the safety height of the adjusting pool, the range cannot be input beyond, and is stored in a database.

[0242] An intelligent pipe network scheduling method driven by big data and based on demand, characterized by using a neural network algorithm to intelligently control the intelligent scheduling of pipe network water supply, and the steps are as follows:

[0243] Establishing a water demand prediction LSTM neural network model

[0244] Collecting the cumulative flow of the effluent of a water supply plant in a water supply area, the liquid level difference of an adjusting pool, weather, week and other data as training input vectors, determining the safety range of the adjusting pool on the pipe network according to experience, establishing a training set for water demand prediction, training the water demand prediction LSTM neural network using the water demand training set to obtain a water demand prediction model, and predicting the water demand using real-time data vectors through the water demand prediction model;

[0245] Establishing a water pump water supply BP neural network model

[0246] Collecting the running condition and power of a frequency conversion pump, water demand, the number of fixed frequency pumps and other data in the water supply area as input vectors, collecting the effluent data of the water plant as output vectors, establishing a training set for water pump water supply prediction including the input vectors and the output vectors, training the water pump water supply BP neural network model using the water pump water supply prediction training set to obtain a water supply prediction model;

[0247] Intelligent scheduling method

[0248] Using real-time water demand data to predict the water demand through the trained water demand prediction LSTM model, using real-time water supply data vectors to predict the water supply through the trained water supply prediction BP model, obtaining an optimal scheduling scheme through an intelligent scheduling algorithm, and regulating and controlling according to the scheme;

[0249] Specifically, the LSTM neural network model: taking an LSTM (Long Short Term Memory neural network, which is a mainstream neural network algorithm suitable for processing and predicting important events with very long intervals and delays in time series) algorithm as the main body, and being composed of a time series neural network through normalization, a 3-layer long short term memory neural network, and denormalization algorithms.

[0250] The characteristic data is normalized, that is, the characteristic data is mapped between 0 and 1 for processing, which is faster and improves processing performance.

[0251] The normalization operation is performed, that is, the characteristic data is mapped between 0 and 1 for processing, which is faster and improves processing performance.

[0252] The first layer LSTM uses C1 (tested, the first layer 128 is the optimal solution) cell units, and the activation function uses Sigmoid (Sigmoid is between 0 and 1, which helps to update or forget data, because any number multiplied by 0 is 0, which will cause the value to be small or be "forgotten").

[0253] The dropout layer is used to solve overfitting, and the dropout probability parameter is set to 0.1 as the optimal solution.

[0254] The second layer LSTM uses C2 (tested, the second layer 64 is the optimal solution) cell units, and the activation function uses Sigmoid.

[0255] The dropout layer is used to solve overfitting, and the dropout probability parameter is set to 0.5 as the optimal solution.

[0256] The third layer LSTM uses C3 (tested, the third layer 32 is the optimal solution) cell units, and the activation function uses Sigmoid.

[0257] The dropout layer is used to solve overfitting, and the dropout probability parameter is set to 0.5 as the optimal solution.

[0258] The full connection output layer is used, and the output parameter is set.

[0259] The inverse normalization operation is performed, that is, the predicted result data between 0 and 1 is restored to the characteristic data through the inverse normalization function.

[0260] The predicted result is output.

[0261] Specifically, the BP neural network model comprises:

[0262] The feedforward neural network is abbreviated as BP, and the BP neural network is a multi-layer feedforward network trained by error backpropagation, which is abbreviated as error backpropagation. The basic idea is the gradient descent method, which uses gradient search technology to minimize the mean square error of the actual output value and the expected output value of the network. Based on the BP neural network, through normalization, BP, and inverse normalization algorithms, a neural network is obtained.

[0263] The feature data is constituted by using one record, multiple column input and single column output, and normalized operation is performed, that is, the feature data is mapped between (0, 1) for processing, so that the processing is faster and the processing performance is improved. The first layer Dense (full connection layer) layer is an input layer, uses D1 (after testing, 64 cells in the first layer are the optimal solution) cell units, and uses a relu (rectified linear unit, which is a commonly used activation function in a neural network) activation function. The second layer Dense (full connection layer) layer is an input layer, uses D2 (after testing, 64 cells in the second layer are the optimal solution) cell units, and uses a relu activation function. The third layer Dense (full connection layer) layer is an output layer, the output layer has only one cell unit, the output column is 1, and inverse normalization operation is performed, that is, the result data between (0, 1) obtained by prediction is restored to feature data through an inverse normalization function, and a prediction result is output

[0264] Specific intelligent scheduling algorithm: used for intelligent scheduling of water pumps according to water supply areas, intelligent calculation according to regulation pool prediction data and water demand prediction data, including regulation pool liquid level prediction algorithm and water pump scheduling algorithm.

[0265] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A big data driven on-demand intelligent pipe network scheduling system, characterized in that: The system comprises an access gateway module, a data standardization module, a database, an AI engine module, and a Portal module. The access gateway module is used to collect real-time data of the Internet of Things intelligent devices, and the Internet of Things intelligent devices are linked to the access gateway module through a network and interact with the access gateway module through a transmission protocol. The data standardization module comprises real-time standardization and timing standardization. The real-time standardization is used to standardize the collected data, and the standardized data are used by the AI engine module for prediction and stored in the database. The timing standardization is used to read the standardized data of each device stored in the database, unify the historical data of each device with different collection frequencies into characteristic data with a frequency D, and store the characteristic data in the database as training data for each model training module of the AI engine. The database is used to store the data of the Internet of Things intelligent devices collected by the access gateway module, weather data, standardized data processed by the data standardization module, prediction data of the AI engine module, data of a water supply area, and data of a regulating pool liquid level range. The AI engine module is used to train a model by calling historical data of a pipe network, predict and analyze real-time data of the pipe network, and provide a scheduling scheme. The water pump water supply and intelligent scheduling module is used to intelligently schedule and predict the water supply of the water pumps in each water supply area. The water demand prediction module is used to predict the water demand of each water supply area according to the water supply area, train a model by using time series neural network algorithm and historical data of the area, and predict the water demand in a future period of time by using the trained model and real-time data. The regulating pool liquid level prediction module is used to read the historical standardized data, train a model by using a multi-layer time series neural network algorithm, and predict the regulating pool liquid level in a future period of time by using the trained model and real-time data. The Portal module is used to display real-time data, historical data, and analysis data of the Internet of Things intelligent devices in each process.

2. The intelligent pipe network scheduling system driven by big data and based on demand for supply according to claim 1, wherein: The uplink message refers to data indicators collected from the Internet of Things intelligent devices and weather data obtained from the Internet. The downlink message refers to a message generated by the intelligent scheduling module and sent to the water pumps through the access gateway module to remotely control the start, stop, and frequency setting of the water pumps. The real-time standardization uses a method of calculating a standard deviation by using historical data in H days, determining that adjacent data with an absolute value exceeding M times of the standard deviation is abnormally high or low, regarding data less than 0 as abnormal data, deleting the abnormal data, and storing the normal data in an intermediate table in the database. The algorithm of the timing standardization unifies the frequency D as follows: The flow is the cumulative flow per D time; The pressure is the average pressure per D time; The pump frequency is the average frequency per D time; The liquid level is the average liquid level per D time; The storage capacity of the regulating pool is the liquid level difference per D time X the bottom area of the regulating pool; The weather is the average temperature per D time; The liquid level difference is the average liquid level per D time - the average liquid level per (D-1) time; The water demand is the flow - the storage capacity of the regulating pool; The weather data is the real-time air temperature data obtained from the Internet.

3. The intelligent pipe network scheduling system driven by big data and based on demand according to claim 2, characterized in that: In the real-time standardization method, H days are 30 days, M times are 3 times, and D is 1 hour; The real-time standardized data includes flow data, pressure data, frequency data, and liquid level data.

4. The intelligent pipe network scheduling system driven by big data and based on demand according to claim 2, characterized in that: The water pump water supply and intelligent scheduling module includes a water pump water supply model training module, a water pump water supply data prediction module, and a water pump intelligent scheduling module; The water pump water supply model training module uses the standardized data as model training data, uses a feedforward neural network to perform model training, saves the trained model in a specified model file directory, and calls the water pump water supply data prediction module to update the used data and perform model training to generate a new model at a specified update time; The water pump water supply data prediction module reads the future N-hour water demand data predicted by the water demand prediction module and the current water pump frequency real-time data in the database, and predicts the future N-hour water output; The water pump intelligent scheduling module intelligently schedules the water pumps according to the water supply area, intelligently calculates based on the regulating pool liquid level prediction data and the water demand prediction data, and is set to execute once every W hours in the program; The water demand prediction module includes a water demand prediction model training module and a water demand data prediction module; The water demand prediction model training module reads the historical data of the water demand, water output pressure, regulating pool liquid level difference, weather, and day data indicators processed by the standardization module in the database, sorts the data by time to form a time series, uses a multilayer time series neural network algorithm to perform model training, saves the trained model in a specified model file directory, and calls the data prediction module to update the used data and perform model training to generate a new model at a specified update time; The water demand data prediction module uses the real-time data of the water demand, regulating pool liquid level difference, weather, and day data indicators processed by the data standardization module, calls the trained model of the water demand prediction model training module, predicts the water demand in a future period of time, and stores the prediction result in the database; The regulating pool liquid level prediction module includes a regulating pool liquid level prediction model training module and a regulating pool liquid level data prediction module. The regulating pool liquid level prediction model training module reads the historical data of the water plant effluent flow, water plant effluent pressure, regulating pool liquid level, weather, and week data indicators processed by the standardization module in the database, sorts the data by time to form a time series, uses a multi-layer time series neural network algorithm to train the model, saves the trained model in a specified model file directory for the data prediction module to call, sets an update time to update the used data, and simultaneously executes model training to generate a new model; The regulating pool liquid level data prediction module uses the real-time data of the water plant effluent flow, water plant effluent pressure, regulating pool liquid level, weather, and week data indicators processed by the data standardization module, calls the trained model of the regulating pool liquid level prediction model training module, predicts the regulating pool liquid level in the future period of time, and stores the prediction results in the database.

5. The big data-driven intelligent pipe network scheduling system according to claim 4, characterized in that: The number of models trained by the water pump water supply model training module is determined according to the type of water pump: In the first case, all are variable frequency pumps, the number L of variable frequency pumps in the water pump information set in the current water supply area management module is read, the pump frequency, water demand data, and pump unit efficiency processed by the standardization module in the database are read as input indicators for model training, L models are established, and models are established according to the number of pumps in the combined pump unit combination; In the second case, all are fixed frequency pumps, the number of fixed frequency pumps under the current water supply area is read, the historical data processed by the standardization module in the database, the number of pump operation, water demand, and pump unit efficiency are read as input indicators, and the water plant effluent quantity is taken as the output; In the third case, all are variable frequency pumps, the number L of variable frequency pumps in the water pump information set in the current water supply area management module is read, the pump frequency, water demand data, and pump unit efficiency processed by the standardization module in the database are read as input indicators for model training, L models are established, and models are established according to the number of pumps in the combined pump unit combination, the number of fixed frequency pumps does not affect the number of models, and the number of fixed frequency pumps is taken as an input indicator; The water pump intelligent scheduling module includes a regulating pool liquid level prediction algorithm and a water pump scheduling algorithm; The regulating pool liquid level prediction algorithm calculates the liquid level in the future N hours according to the future N hours of water supply predicted by the water supply data prediction module and the future N hours of water demand predicted by the water demand data prediction module, and the algorithm is as follows: Regulating pool predicted water storage = sum of predicted future N hours of water volume - sum of future N hours of water demand; Regulating pool predicted liquid level difference = regulating pool predicted water storage / regulating pool bottom area; Future N hours of liquid level = current regulating pool real-time liquid level + regulating pool predicted liquid level difference The water pump scheduling algorithm sets the whole hour time i within the future N hours, adjusts the frequency of the variable frequency water pump or the number of the fixed frequency pump, re-calls the water pump water supply data prediction module and the water demand data prediction module to respectively perform water supply prediction and water demand prediction, the algorithm is used to predict the adjustment pool level change in the future N hours according to the scheduling of the water pump, to adjust the pool level to be within the safe liquid level range before i reaches N, to generate the device operation instruction at the same time, and to send the device operation instruction to the device; The algorithm is as follows: First step: judging whether the adjustment pool level is within the reasonable range according to the current pump operation condition Using the current pump operation condition and the water demand predicted by the water demand data prediction module as the input, calling the water supply predicted by the water pump water supply data prediction module, calling the adjustment pool level prediction algorithm to calculate the adjustment pool level in the future N hours, and judging whether it is within the reasonable range; If it is not within the reasonable range, proceed to the second step; If it is within the reasonable range, return the scheduling information; Second step: selecting the pump to be operated When the number of the current running pump is 1, selecting the current pump as the pump to be operated; When the number of the current running pump is greater than 1: calling the efficiency priority algorithm to calculate and select the pump to be adjusted; Third step: operating according to the selected pump type First, the selected pump is a variable frequency pump: adjusting the frequency of the selected pump according to the step P; When it exceeds the range, lower the pump frequency; when it is lower than the range, increase the pump frequency; Adjusting the frequency of the selected pump, increasing or decreasing the pump frequency by P to obtain the adjusted pump operation condition, setting i = 1, i <= N, starting from the i hour, using the current pump operation condition and the water demand predicted by the water demand data prediction module as the input before the i hour, using the adjusted pump operation condition and the water demand predicted by the water demand data prediction module as the input after the i hour, calling the water pump water supply data prediction module and the water pump intelligent scheduling module N times, the adjustment pool level prediction algorithm calculates the adjustment pool level in the future N hours, and judges whether it is within the reasonable range; if it is still not within the reasonable range, continue to increase or decrease the frequency of the selected pump by P, and then judge whether it is within the reasonable range, each time increasing or decreasing P until it is within the reasonable range, and returning the scheduling information; If the following two situations occur during the adjustment process, they are handled as follows: The frequency of the selected pump is adjusted to or below the minimum value, the selected pump needs to be stopped, and the first step is returned to operate; The frequency of the selected pump is adjusted to or above the maximum value, the second step is returned to increase one pump; if the number of the current pump running is 0, the scheduling information is returned; if all the pumps are in the running state and the frequency of each pump is adjusted to or exceeds the maximum value, the scheduling information is returned; Second, the selected pump is a fixed frequency pump: adjusting the number of the running pump, when it exceeds the range, stopping the selected pump, when it is lower than the range, starting the selected pump; The number of pumps after stopping or starting is used as input, i=1, i<=N. Starting from the i-th hour, the current operating status of the pumps before the i-th hour and the water demand predicted by the water demand data prediction module are used as input. After the i-th hour, the operating status of the pumps after adjustment and the water demand predicted by the water demand data prediction module are used as input. The water supply data prediction module and the intelligent pump scheduling module are called N times in a loop. The regulating tank level prediction algorithm calculates the regulating tank level for the next N hours and determines whether it is within a reasonable range. If it is within a reasonable range, the scheduling information is returned. If it is still not within a reasonable range, the second step is returned. If the number of operating pumps is 0 or all are running, the scheduling information is returned. The fourth step is to generate device operation instructions and send them to the device; Based on the returned scheduling information, a downlink control command is generated and remotely controlled through the access gateway module. The downlink control command is sent to the pump that needs to be remotely controlled after adding i hours to the current actual time, and the pump is started / stopped. Downlink control commands are messages sent to the water pump that can be recognized and executed by the water pump. The specific format is provided by the water pump manufacturer or the communication module manufacturer on the water pump. The efficiency-first algorithm, based on historical pump data, calculates the pump unit efficiency and prioritizes high-efficiency combinations. When there are E pumps in operation and F pumps not in operation, the following scenarios apply: When screening requires adjusting pumps: combine any X pumps from the running pumps, call the pump unit efficiency algorithm, sort the individual pump efficiency of the X pumps, and traverse in order to select the pumps that have not been adjusted to the highest or lowest efficiency. When screening pumps that need to be shut down: combine any X pumps from the running pumps, call the pump unit efficiency algorithm to calculate the unit efficiency, retain the combination with the highest unit efficiency, and the remaining 1 pump is the selected pump if it is not among the retained units, X=E-1; When screening for pumps that need to be added: combine all the pumps in operation with the pumps that are not in operation, i.e., Z pump combinations, calculate the unit efficiency for each combination, and retain the combination with the highest unit efficiency. Among them, the pumps that are not in operation are selected as the pumps, Z=E+1. The efficiency algorithm for the water pump unit: Output power = (instantaneous flow rate x 1000 / 3600) x (pressure x 100 - level in regulating tank); Single pump input power = voltage x current x 1.732 x 0.95; Unit pump output power sum = Σ single pump input power; Pump unit efficiency = output power / sum of pump output power.

6. A big data driven on-demand smart utility network dispatch system as claimed in claim 1 wherein: The Portal module includes a water intake pumping station module, a clear water tank module, a water delivery pumping station module, a regulating tank module, a today's water supply status module, a source water level module, a pipeline data module, a water quality monitoring module, a data analysis module, a water supply area management module, and a regulating tank level range setting module. The water intake pump station module is used to display real-time and historical data on the operation of the water intake pump. Data indicators include operating status (operating / stopping), operating / stopping duration, rated flow rate, and frequency. It can also select a time range to query historical data of each indicator for the selected time period from the database. The data is read from the database. The clear water pool module is used for displaying real-time data and historical data of a liquid level instrument, and data indexes include a liquid level, and historical conditions of the liquid level can be inquired from a database according to time, and the historical conditions are displayed on a page in a curve form, and the data are read from the database; The water supply pump house module is used for displaying real-time data and historical data of a running condition of a water supply pump, and data indexes include a running condition (running / stop), a running / stop time length, a rated flow, and a frequency, and historical data of each index in a selected time range can be inquired from the database according to a time range, and the data are read from the database; The today's water supply condition module includes a today's water demand, a today's water supply, and a today's water storage, and is used for displaying and calculating the today's water demand, the today's water supply, and the today's water storage, and the data are read from the database, and a calculation method is as follows: The today's water demand = the today's water supply - the today's water storage; The today's water supply refers to a cumulative flow of water plant outflow from 0 o'clock to a current time; The today's water storage refers to a water demand of a regulating pool from 0 o'clock to a current time; The source water module is used for displaying real-time data and historical data of a source water liquid level instrument on a page, and data indexes include a source water liquid level, and historical conditions of the source water liquid level can be inquired from a database according to time, and the historical conditions are displayed on the page in a curve form, and data are read from the database; The pipe network data module reads pressure data and position data of pipe network key nodes collected by the gateway module from a database, the position data refers to longitude and latitude information of an installation position of a pressure equipment, and is displayed on a page, and historical conditions of pressure of each pipe network key node can be inquired from the database according to time, and is displayed on the page, and historical maximum pressure and minimum pressure thresholds are found out by analyzing historical data of pressure of the pipe network key nodes, and real-time data are analyzed for abnormal condition alarm, such as high pressure and low pressure alarm, and when alarm of a certain pipe network node is generated, the pipe network node is displayed in red on the page, and high pressure or low pressure alarm is prompted; The water quality monitoring module is used for displaying real-time water quality data and historical data trends of each process section on a pipe network, and includes a water quality module, a turbidity trend module, and a residual chlorine trend module; The water quality module reads real-time data of turbidity, PH value, and residual chlorine water quality related indexes of main process sections from a database, and displays the real-time data on a page, and process sections displayed on the page include raw water quality, sedimentation tank water quality, filtered water quality, and finished water quality; The turbidity trend module reads historical data of turbidity of the sedimentation tank, filtered water, and finished water process sections from a database, and displays turbidity trends of the process sections on a page; The residual chlorine module reads historical data of residual chlorine of the filtered water and finished water process sections from a database, and displays residual chlorine trends of the process sections on a page; The data analysis module is used for analyzing prediction accuracy of water supply prediction, water demand prediction, and regulating pool liquid level prediction, and includes a water supply prediction analysis module, a water demand prediction analysis module, a clear water pool prediction analysis module, and a regulating pool liquid level prediction analysis module; The water supply prediction analysis module reads the water supply area from the database, reads the water supply prediction data and the standardized historical data from the database according to the water supply area, displays the comparison curve through the chart, and views the fitting condition of the two curves; the prediction data of the future N hours and the prediction accuracy of each hour are displayed; The water demand prediction analysis module reads the water supply area from the database, reads the water demand prediction data and the standardized historical data from the database according to the water supply area, displays the comparison curve through the chart, and views the fitting condition; the prediction data of the future N hours and the prediction accuracy of each hour are displayed; The regulating pool liquid level prediction analysis module reads the water supply area from the database, reads the regulating pool liquid level prediction data and the standardized historical data from the database according to the water supply area, displays the comparison curve through the chart, and views the fitting condition; the prediction data of the future N hours and the prediction accuracy of each hour are displayed; The prediction accuracy algorithm is: Prediction accuracy = Σ prediction accuracy of each time point / number of time points; The water supply area management module is used for creating a plurality of different water supply areas according to the actual pipe network water supply condition, the area name cannot be repeated, and when the water supply area is created, the information under the area is set, including: Water supply plant information, including identification and name; Regulating pool information, including regulating pool identification and name, regulating pool safety height, regulating pool bottom area; Water pump information, including water pump identification, name, whether frequency conversion, maximum frequency, minimum frequency; The regulating pool liquid level range setting module is used for setting the regulating pool liquid level safety range in different seasons, including: the start time and the end time of each season, a plurality of time periods are set in each season, and the lowest value and the highest value of the liquid level corresponding to each time period; according to the characteristics of the seasons, 1 day is divided into several time periods of water supply peak and valley, the start time and the end time of each time period are input in the time format HH:MM, a plurality of time periods can be added, the time periods cannot be crossed and cover 0-24 hours, a reasonable safety liquid level range value is set for each time period, the lowest value of the liquid level is greater than 0, and the highest value of the liquid level is less than the safety height of the regulating pool, and the range cannot be input, and is stored in the database.

7. A big data driven on-demand smart utility network dispatch system as claimed in claim 5 wherein: N is 6; the update time is set to 0 o'clock every day in the model training of the AI engine module; W is 1; and P is 3.

8. A big data driven on-demand smart pipe network scheduling method, characterized in that Using a neural network algorithm, the pipe network water supply intelligent scheduling is intelligently controlled, and the steps are as follows: Establishing a water demand prediction LSTM neural network model; Collecting the cumulative flow of the water supply plant outflow, the regulating pool liquid level difference, the weather, and the week data in the water supply area as a training input vector, determining the regulating pool safety range on the pipe network according to experience, establishing a water demand prediction training set, training the water demand prediction LSTM neural network to obtain a water demand prediction model, and predicting the water demand by using a real-time data vector through the water demand prediction model; Establishing a water pump water supply BP neural network model; The operation condition and power of a frequency conversion pump in a water supply plant in a water supply area, water demand, and the number of fixed frequency pumps are collected as input vectors, and the water output of the water plant is collected as an output vector, a training set for water pump supply prediction including the input vectors and the output vector is established, and the water pump supply prediction training set is used to train the water pump supply BP neural network model to obtain a water supply prediction model; Intelligent scheduling method; Using real-time water demand data, the trained water demand prediction LSTM model is used to predict the water demand, using the real-time water supply data vector, the trained water supply prediction BP model is used to predict the water supply, the optimal scheduling scheme is obtained through the intelligent scheduling algorithm, and the scheme is adjusted and controlled.

9. The big data driven intelligent pipe network scheduling method according to the needs of claim 8, characterized in that: The LSTM neural network model: The first layer LSTM uses 128 cell units, uses a dropout layer to solve overfitting, and sets the dropout probability parameter to 0.1; The second layer LSTM uses 64 cell units, uses a sigmoid activation function, uses a dropout layer to solve overfitting, and sets the dropout probability parameter to 0.5; The third layer LSTM uses 32 cell units, uses a sigmoid activation function, uses a dropout layer to solve overfitting, and sets the dropout probability parameter to 0.

5.

10. A big data driven on-demand smart utility network scheduling method as claimed in claim 8, wherein: The BP neural network model, the first layer Dense layer uses 64 cell units, and the activation function uses relu; the second layer Dense layer uses 64 cell units.

Images