CN110570307B - Whole market multi-variety gold financing management system based on intelligent strategy platform - Google Patents

Abstract

The invention discloses a market-wide multi-variety financial resource management system based on an intelligent strategy platform, which comprises: the transaction platform is responsible for receiving the newspaper command, executing the algorithm transaction and interfacing the market newspaper function; the data platform is responsible for service quotation access and production of related derived quotations; the strategy platform is responsible for the online operation of strategies for strategy production and high performance, comprises an offline strategy production management subsystem and an online strategy operation management subsystem, and adopts a machine learning algorithm to calculate the weight of the fund proportion of each strategy in the combination, and selects investment combination weights in each period so as to maximize the final benefit; the policy platform comprises an online policy container cluster, a policy platform and a policy platform, wherein the online policy container cluster supports single-node multi-container and horizontal expansion; and the online operation link adopts a CEP event processing engine and is constructed on an Apama container to carry out the circulation of core business.

Description

Whole market multi-variety gold financing management system based on intelligent strategy platform

Technical Field

The invention belongs to the technical field of financial information, and particularly relates to a full-market multi-variety gold financing system based on an intelligent strategy platform.

Technical Field

At present, the financial industry has entered the large asset management era, the scale of asset management reaches the amount of days, and simultaneously, the competition barrier between the financial homonymies is broken, on one hand, different financial institutions are allowed to develop homogeneous asset management business, on the other hand, the extension of original system business is continuously expanded, the connotation is continuously abundant, and the business synergy of financial license plates is obviously enhanced. Asset management is a typical knowledge-intensive, talent-intensive industry, highly dependent on human capital with expertise and capabilities. The financial information system business has wide involved area, strong specialization and complex structure, and the outside needs to get through barriers among links with different fields, different institutions and different values; the barrier among the product interface, the service interface and the organization interface is required to be opened inside; in terms of products and services, competitive products need to be designed, different pricing needs to be implemented on products and clients with different risk degrees, and risks of different types, attributes and features are scientifically and effectively segmented and recombined so as to achieve reasonable matching between risks and benefits. The processing, analysis and decision of massive financial data are not available by simple manpower, and the financial industry needs more intelligent and automatic information tools, so that the dependence on manpower is reduced, and the efficiency is improved. Thus, a need exists for a close integration of advanced, secure and efficient information technology systems with asset management processes that lead to innovations in the asset management industry.

Disclosure of Invention

The invention aims to provide a full-market multi-variety gold financing management system based on an intelligent strategy platform, which is used as an intelligent quantitative transaction platform and can be used for acquiring, processing and storing financial market data, producing, detecting back, optimizing and storing strategies, arranging, deploying, operating and controlling wind and managing strategy examples, further obtaining transaction targets, transaction signals and excellent net value curves, and realizing full-dimension intelligent resource management service.

The technical proposal is as follows:

a market-wide multi-variety financial resource management system based on an intelligent policy platform, comprising:

the transaction platform is responsible for receiving the newspaper command, executing the algorithm transaction and interfacing the market newspaper function;

the data platform is responsible for service quotation access and production of related derived quotations;

the strategy platform is responsible for the online operation of strategies for strategy production and high performance, comprises an offline strategy production management subsystem and an online strategy operation management subsystem, and adopts a machine learning algorithm to calculate the weight of the fund proportion of each strategy in the combination, and selects investment combination weights in each period so as to maximize the final benefit; the policy platform comprises an online policy container cluster, a policy platform and a policy platform, wherein the online policy container cluster supports single-node multi-container and horizontal expansion; and the online operation link adopts a CEP event processing engine and is constructed on an Apama container to carry out the circulation of core business.

Preferably, the offline policy production management subsystem comprises 3 sub-modules: the strategy production module is used for realizing the functions of flow control, report display and list query; the strategy return testing optimizing module is used for realizing the establishment and operation of strategy return testing tasks, the scheduling of parallel computing tasks and the injection of strategies into a return testing container; the return container module is used for running the strategy during return, sending quotation to the strategy, calculating funds and holding a warehouse during running the strategy and providing strategy API common strategy calling; and during the operation of the online policy operation management subsystem, the factor calculation module establishes a network link with the return container, receives a policy signal from the return container, calculates a policy evaluation factor and a scoring single curve according to the signal, and calculates comprehensive scoring comprehensive curves of policies on a plurality of varieties after all policy examples are operated.

Preferably, the policy platform comprises produce, optimize, factor and strategy-api modules, and the interaction principle of each module is that an upstream module calls a downstream module to adopt a dobbo service, and the downstream module informs the upstream module of adopting an MQ mode; the product module is responsible for image evaluation main flow control, service inquiry and metadata management, and the image return time is related in the image evaluation process, and the image return is realized by dubbo calling optimize; the optimize is responsible for the return measurement of images and strategies, optimization, task scheduling of return measurement, acquisition of containers by dubbo call infracuction, strategy injection and strategy initial instruction sending; the Infrastructure is responsible for interacting with the policy container, including the management of the online policy container and the offline policy container; the strategy-api is responsible for online policy, and starting and stopping of online policy; the factor is responsible for accepting orders from the offline container corelocker at the time of the back test, and calculating policy factors and scores, the completion of which is notified to the optimize or product module by sending an MQ message.

Preferably, the policy platform comprises a cluster of retested containers: the cluster comprises 6 machines, each node deploys a container process to support horizontal expansion; the policy platform includes a factor computation cluster: the single machine multi-instance deployment, wherein one deploys 4 services, the other deploys 2 services, and the number of services and the number of the callback containers are 1:1.

Preferably, the image evaluation process of the policy platform: the Manager receives an image evaluation task starting request and transmits the request to the product module through a dubbo call; the product starts image evaluation and optimizing step by step, and each image of each layer needs to call optimize for back measurement; the Optimize receives the image return request, creates an image return task, realizes parameter frying through permutation and combination, and starts the return task.

Preferably, the starting and testing tasks of the policy platform include: acquiring available policy container resources, injecting policies into a container, and sending a starting instruction to the container; when the strategy in the container back measurement enters the operation, an order signal is sent to a Factor, and the Factor receives the order to start calculating the strategy performance; after the Factor calculates the policy performance, a report of completion of the test is sent back to the Optimize, and the Optimize performs task ending processing according to the report, including modifying task state bodies, and releasing container resources.

Preferably, the policy container of the policy platform is asynchronous event driven process: if the environment is the return environment, directly reading a local Sqlite file through a monitor in the container to send quotation; if the online container is an online container, subscribing to quotations from the Data-Source; asynchronous event delivery: quotation/signals are circulated in the policy container in an asynchronous event mode; the online strategy supports the market subscription and driving of the complete Bar, currentBar, tick full period and full variety; the strategy internal reading history market data adopts a local memory cache to reduce a large number of IO operations, and the strategy running time is compressed to microsecond level; after the strategy in the container generates the signal, whether the signal is back-measured or on-line is judged according to the environment variable, if the signal is a back-measured task, the signal is sent to the Factor module for performance calculation, and if the signal is an on-line container, the signal is directly sent to the transaction module and then forwarded to the transaction market.

Preferably, the under-fitting is solved: the method comprises the steps of carrying out back testing aiming at the whole market, selecting a variety with better fluidity for carrying out back testing, and avoiding the problem of under fitting through image evaluation and strategy screening; solution of overfitting: in the process of image evaluation, the maximum layer number of the strategy image is limited to 3, plain parameters are adopted, parameter islands are limited, and in addition, a separate test between the inside and the outside of the sample is adopted, the strategy is trained in the sample, and the outside of the sample is tested in a return mode to verify the validity of the strategy generated in the sample.

Preferably, the automated production process of the strategy: creating an overall scheduling task in the product, receiving page data, storing mongo, then starting dimension element arrangement, after the task is successful, sending an MQ notification to the product module, and if the execution of a certain step fails, automatically constructing the overall task to be stopped.

Drawings

FIG. 1 is a simplified logic architecture diagram of a muji system.

FIG. 2 is a physical architecture deployment diagram of the online running portion of the muji system.

Fig. 3 is a flow chart of a transaction statement.

Fig. 4 is a flowchart of a bin and month shift.

Figures 5, 6, and 7 are flow charts of the newspaper, drop and check of the transaction microservice subsystem, respectively.

Fig. 8 is a flowchart of automatic bin difference monitoring.

FIG. 9 is an algorithm single process flow diagram.

FIG. 10 is a diagram of a data platform hardware deployment architecture.

FIG. 11 is a functional flow diagram of a data platform.

Fig. 12 is a business flow diagram of a data platform.

Fig. 13 is a logical block diagram of a policy platform.

FIG. 14 is an image evaluation timing diagram of the policy platform.

Fig. 15 is a policy container-asynchronous event driven flow diagram.

FIG. 16 is a schematic diagram of a policy platform parallel computing task scheduling.

Fig. 17 is a policy screening flow chart.

Fig. 18 is a strategy automated production flow diagram.

Fig. 19 is a schematic diagram of a strategic production parallel computing architecture.

Fig. 20 is a diagram of a strategically operated linear expansion architecture.

Detailed Description

In order to enable those skilled in the art to better understand the technical scheme of the invention, the full-market multi-variety gold financing system based on the intelligent strategy platform provided by the invention is described in detail below with reference to the embodiment. The following examples are only illustrative of the present invention and are not intended to limit the scope of the invention.

The market-wide multi-variety gold financing system based on the intelligent policy platform in this embodiment is named a mu chicken (muji) system. Full market concepts include domestic and foreign markets, and trade of multi-variety financial products including, but not limited to, foreign exchange, futures, stocks, bonds, and digital currency products.

As shown in fig. 1, the pheasant (muji) system mainly comprises three parts: the transaction platform is responsible for receiving the newspaper command, executing the algorithm transaction and interfacing the market newspaper function; the data platform is responsible for service quotation access and production of related derived quotations; and the strategy platform is responsible for strategy production and high-performance strategy on-line operation. As shown in fig. 2, the whole online running link of the wood chicken adopts a CEP event processing engine, and the CEP event processing engine is built on an Apama container to circulate core business. The data platform provides a service of quotation pushing and a function of basic surface data/history quotation inquiry; the strategy online container receives market data, and the strategy executes strategy ideas based on market events to generate transaction signals; when the transaction container receives the transaction signal, a series of logic processes such as generating a mother bill, removing the bill by an algorithm, and finally reporting the order to the market; after market trading, a return event is pushed to a trading channel, then the return event enters a trading container, and order data is maintained; and finally, finishing calculation of the holding and fund correlation. Table 1 shows the development environment and the running environment of a Muji (muji) system.

Table 1

I. A transaction platform: the transaction platform takes on multiple roles of transaction signal processing, algorithm splitting, market docking, fund calculation and risk control in the wood chicken system. And supporting an online transaction part of the whole wood chicken system, completing algorithm transaction, risk control and core functions of market route distribution. High security, namely, fund security, data security, policy security and system security; the method is high in availability, single-point faults are avoided, account isolation and strategy isolation are realized; high performance, namely data acquisition performance, data processing performance, policy execution performance and high concurrency performance; high accuracy, namely accuracy of data processing, accuracy of transaction signals and accuracy of fund management; the system is stable, can run for a long time without faults, if any, and can recover in a very short time through a perfect fault recovery mechanism. Supporting the whole market, namely supporting the online transaction of a financial target in the domestic market and the overseas market; the whole variety is a multi-level financial trade target for supporting stocks, futures, options, bonds, foreign exchange, funds and the like; full-automatic, based on artificial intelligence technology, adopting a mode of combining programmed transaction and algorithm transaction to complete all transaction flows; and the total funds can be freely adapted to products with any funds and output an excellent net value curve, so that the goal of the funds management and value increasing service is realized. Table 2 shows the modules of the trading platform.

Table 2

Transaction noun:

policy instance: a policy instance is a specific execution of a policy (policy combination), essentially a diverse combination of policy subunit instances, which includes three states to be started, run, and terminated.

Transaction unit-in one strategy (strategy combination) example, the whole process from warehouse building to empty warehouse of a single transaction variety is called a transaction unit, and the middle part comprises warehouse adding, warehouse subtracting and the like.

And (3) moving the warehouse to change months, namely switching futures main force contracts, and carrying out the processes of leveling old contracts and opening new contracts on contract real discs in the wooden chicken system.

And a strategy bin, namely accumulating and generating positions held by a strategy instance theory according to strategy signals.

And (3) a real disc bin, namely, actually holding positions of the current strategy example, and accumulating according to the sub-sheet transaction returns.

The bin difference is the difference value between the strategy bin and the real disc bin, and the bin difference=strategy bin-real disc bin; the bin difference is positive, and the sent entrusting instruction is consistent with the direction of the strategy signal instruction; when the bin difference is negative, the direction of the issued entrusting instruction is opposite to that of the strategy signal instruction.

And (3) ending the algorithm list, namely indicating that the parent list is in a completion state, and reporting all the child lists (including success in forming, list removing or list refusing).

Automatic alignment, which is to automatically align the position of the real disc bin with the position of the strategy bin in the process of executing the transaction signal.

Non-automatic alignment-corresponding to an automatic alignment task, as distinguished from manual alignment. Refers to a transaction mode in which an algorithm is adopted to conduct automatic commission transactions without bin alignment.

Manual alignment, as distinguished from automatic transactions (including automatic alignment versus non-automatic alignment), refers to a transaction behavior in which manual intervention employs a form of manual intervention to conform a policy cartridge to a real disc cartridge. The manual alignment function is active in policy instance running states (including running and suspended).

And the price tracking list is that when the order list is not submitted for a period of time, the order list is removed, the new order list is adopted to re-report the order list (the price tracking list), and the price of the newly reported order list can be divided into the price tracking list, the price tracking list and the latest price tracking list. Sometimes the price pair is also referred to as an opponent price.

Price-matching and list-following: if buying is to be performed, buying is performed by using a selling price for price tracking.

Hanging price and tracking list: if buying is to be carried out, buying is carried out by using buying price after the price is hung.

The latest price list is tracked: the latest price chasing list is bought or sold using the latest price (LastPrice).

Price limiting instruction: the order is reported by a limit order specified by the transaction, and the order needs to specify the order price.

Market price instruction: the market price specified by the exchange is used for reporting, the order does not need to specify the order price, and the exchange system can automatically convert the market price instruction into the price of the expansion and reduction stop plate of the order direction to participate in the transaction. The buying market price list can be converted into the rising market price to buy, and the selling market price list can be converted into the falling market price to sell.

FOK: all deals immediately or automatically revoked, meaning that all deals must be at the specified price or else automatically revoked by the system. The result is either full commit or full undo.

FAK: the automatic withdrawal of the remaining quantity of immediate deals refers to the fact that at a specified price, the remaining orders are automatically withdrawn by the system.

Transaction day: domestic futures currently follow a daily settlement system, except for Zhongjin, all are 15:00 off-line in the afternoon, 15:15 on-line in Zhongjin are closed, and settlement is carried out after the off-line. The trade date refers to a trade time period belonging to the same settlement, and the night plate trade date and the natural date are not the same. Such as a night dish for 30 days of 5 months (thursday) night, which should be attributed to a trade day of 5 months and 31 days.

And (3) a: and (5) opening the warehouse in the current transaction day, and holding the warehouse. Yesterhouse: and (5) opening the warehouse in the non-current transaction day, and holding the warehouse. And (3) plain: and (5) leveling the holding of the open cabin in the same day.

Plain yesterday: the open-warehouse of non-current day is carried out to open-warehouse, the current domestic futures institute and Zheng Shang institute, the medium-gold institute does not distinguish the present warehouse, the yesterday warehouse can use the open-warehouse to open-warehouse. The last period and the last energy distinguish the yesterday bin of the present bin, and the present bin must be used for the present bin to be flat.

The application type is divided into a set period protection value (set protection for short), set benefit and an application machine, and generally, all transactions belong to the application machine; the set period guard is mainly used for production and operation of some enterprises; the arbitrage is mainly used for arbitrage trading strategies of some institutions, and some exchanges (such as Zhongjin) can relax the regulatory requirements of the maximum withdrawal number for arbitrage traders.

Contract price minimum variation unit the contract price minimum variation unit refers to the number of points per minimum price variation, for example, the corn minimum variation unit is 1, the coke minimum variation unit is 0.5, and the IF minimum variation unit is 0.2.

Hop count: for the minimum variation unit, for example, the price corresponding to 5 hops of corn is 5*1 =5; coke 5 skip corresponds to a price of 5 x 0.5=2.5; IF5 hops correspond to a price of 1.

In the future market, traders only need to pay a small amount of funds according to a certain ratio of the price of the future contract as a financial guaranty for fulfilling the future contract, and can participate in the buying and selling of the future contract, and the funds are the future guaranties.

Freezing the deposit, namely when the warehouse-opening order report is not yet finished, the collected deposit is called as the frozen deposit.

As shown in fig. 3, the transaction report flow of the transaction container in the transaction platform is as follows: s1: after the transaction signal reaches the transaction container, the signal initialization and the strategy bin association are realized through logic processing of the signal management module; s2: after the signal is dequeued from the signal management module, the signal enters the order management module for filtering processing to generate a mother bill; s3: the master list entering algorithm module processes the parallel and parallel instruction through the algorithm route, and then routes the parallel and parallel instruction to the corresponding list splitting algorithm to split the list, and after all the hand numbers split the list, a signal for finishing the list splitting is generated for the master list to the algorithm post module; s4: the sub-bill generated in the bill disassembling process is directly sent to a fund checking module to carry out wind funding and fund checking logic, and the sub-bill is sent to a market route after checking, otherwise, the fund checking refuses the bill; s5: the transaction routing module defaults to send a request to the transaction channel gateway in a java plug-in mode, and sends the request to a corresponding transaction market through routing; s6: the return sent back from the market is received by the client side built in the transaction channel, then the basic service of container management is called to send the return back to the transaction container, the return entering the transaction container is received by the sub-bill state monitor and triggers corresponding state circulation, the return is submitted to the warehouse, the state processing of withdrawing/rejecting the bill is carried out, and finally the state promotion of fund calculation and mother bill is triggered.

In a transaction report flow S2 of a transaction container in a transaction platform, an order management module performs three-layer filtering: processing signals of the type under the master force mapping to obtain corresponding master force contracts; the bin alignment, the logic of automatic alignment or non-automatic alignment of bin is processed to correct the number of single hand of newspaper; and (3) carrying out secondary confirmation of the strong flat type, and if the secondary confirmation is not consistent with the real disc bin, re-initializing the number of single hand of the report, and finally generating a mother sheet. After the transaction report flow S6 of the transaction container in the transaction platform, the transaction report triggers to perform fund calculation: the method comprises the steps of calculating layer by layer in the form of java plug-in, namely firstly holding the warehouse and fund of the warehouse holding dimension of the strategy instance dimension, then fund of the strategy instance dimension, and finally fund of the account dimension. After the transaction report flow S6 of the transaction container in the transaction platform, the state of the mother bill is pushed along with the corresponding action of the mother bill, and if the mother bill is ended, the result is fed back to the signal management module to complete the pushing of the signal state.

The transaction signal received by the transaction container in the transaction platform is initially INIT after the signal management module of the container enters a warehouse; each transaction signal after warehousing is positioned in a logic queue, and the logic main key of the queue is stratInstId+symbol+position direction; after non-trade time or high-priority signals are enqueued, part of signals are directly pushed to an UNDO state; if the same signal types are in INIT, merging, and pushing the MERGED signals to MERGED; the signal is normally dequeued for execution, and the state flow is transferred to WAITING WAITING algorithm for execution; the signal enters the algorithm to execute the sheet disassembly, and the state is changed into RUNNING; after the signal in the RUNNING state is terminated by the signal of high priority, the stream is converted into UNDO; the abnormal condition advances to unknown; normally the signal is DONE to DONE.

As shown in fig. 4, the bin and month shifting process of the transaction platform is as follows: s1: the trading container monitors a bin shifting and month shifting signal S2 sent by a trading timing task system: if the execution signal with low priority exists, the bin shifting and month changing signal can terminate the execution signal with lower priority than the execution signal, and then enter algorithm execution after receiving termination return, otherwise directly enter algorithm. S3: after algorithm transaction is entered, algorithm calculation is started, and if the number of hands is 0, a strategy bin is directly updated to be a new master contract; if the number of hands is greater than 0, a newspaper is sent according to the buying and selling amount and the price, and monitoring of the return of the current order is established. After the warehouse moving and month changing process S3 of the transaction platform, if the return is successful, the transaction record is updated; if the return fails, a request for removing the list is sent; if the report refuses the bill, triggering the wind control alarm monitoring mechanism and sending an alarm signal.

Figures 5, 6, and 7 are flow charts of the newspaper, drop and check of the transaction microservice subsystem, respectively. The transaction platform comprises a transaction micro-service subsystem, and the newspaper flow is as follows: s1: dynamically routing the request to a corresponding transaction channel routing gateway at the client signaled in the market routing module of the transaction container; s2: the channel route receives the report request, carries out secondary service route and forwards the request to the CTP channel or the IB channel; s3: after receiving the request, the corresponding channel firstly checks whether the channel is established, if not, the channel is initialized and established; s4: checking the login state of the user, and if the user does not login, re-login is performed; s5: converting parameters, converting to parameters required by corresponding markets, and converting the newspaper request to a redis triggering asynchronous database; s6: and finally, sending the request to the market and waiting for the subsequent return of the market. After receiving the market return, the report flow S6 of the transaction micro-service subsystem firstly acquires order information from the redis, then constructs an event corresponding to the transaction container, and calls the basic service of container management to send the return to the transaction container.

The form removing operation of the transaction micro-service subsystem comprises the steps of removing a form when a transaction container is not in contact, manually removing a manual form from a page, or removing the form after a disc is timed, wherein the form removing process comprises the following steps of: s1: after receiving the request for removing the list, the OTC forwards the request to a corresponding CTP channel or IB channel according to the corresponding brookereID; s2: carrying out data removal packaging after parameter verification; s3: checking the login state of the user, and if the user does not login, re-login is performed; s4: after obtaining account login connection, directly sending a form removing request; s5: and after synchronously collecting the return of the withdrawal list, carrying out subsequent state change.

The transaction micro-service subsystem sends out a bill overtime or removes a bill overtime, triggers a bill checking service, and the bill checking flow is as follows: s1: firstly, after receiving a list searching request, the OTC forwards the request to a CTP channel or an IB channel according to a corresponding list number; s2: assembling parameters necessary for the list searching request from the cache and the database; s3: checking the login state of the user, and if the login state is not available, performing re-login; s4: and returning the query result.

Suspension and resumption of policy instances of the transaction microservice subsystem may be interfaced by a background administrator. If it is a pause function: firstly, the hen management platform initiates a pause event request, verifies the state of the policy instance through the transaction API service, and sends the verification to the transaction container through the basic API service. The transaction container performs corresponding business execution according to whether the front end selects the automatic bin leveling function or not: if auto-leveling bin is selected, jiang Ping processing is initiated. If no auto-leveling bin is selected, the policy instance is directly paused. If it is the recovery function: firstly, a hen management platform initiates a recovery event request, and the recovery event request is sent to a container to start operations such as bin shifting, month shifting and the like after passing through a transaction API service and then passing through a basic API service.

Bin monitoring operation of transaction microservice subsystem: obtaining bin difference by acquiring strategy bin data; by acquiring available transaction containers and transmitting a transaction signal pair Ji Cangwei; the cause of the bin difference is obtained by obtaining the sub-list of the non-intersection or the partial intersection from the time range.

The wind control system of the transaction platform monitors four risk situations outside the monitoring index range in the execution process of each strategy instance signal in real time: there are real-time signals, four types per minute, per hour, per day: the wind control system checks the opening bill, and if the abnormal behavior supervision index of the example of the opening bill reaches the intervention, the report is refused; the 2-minute timing task and the transaction return in the disk trigger a wind funding risk calculation interface of the strategy instance and the account, the wind funding risk level is calculated, and if the strategy instance level wind funding risk level reaches intervention, a track-api interface is called to pause and level the strategy instance; before the transaction container reports the bill, the wind control plug-in is called to calculate whether the dimension available wind fund of the strategy instance is sufficient, and if not, the bill is directly refused; the timing task system scans the strategy instance and the account dimension risk index monitoring table and alarms the strategy instance and the risk item of which the account dimension reaches reminding, early warning and intervention levels; and the wind control system accumulates early warning values of strategy examples and account dimensions for the withdrawal consignment return and the delivery return, and triggers an alarm after reaching an early warning level.

The transaction platform comprises a timing task subsystem, and the timing task settles on the same day at 8 points 35 before the disc at night, and mainly counts each fund data and guarantees normal transaction on the next transaction day. Because it is possible that the counter does not pay a daily account for a certain day and cannot check the account, the daily account fails on the same day, but the daily account is automatically supplemented by the task later and correctly executed. The flow is as follows: s1: in order to ensure the performance of the day knot, slicing is carried out aiming at the account, and the day knot is carried out as an entrance; s2: inquiring all strategy examples of non-INIT states under the account as a day knot condition; s3: making a matching composite disc by using the warehouse holding record of the day before the successful result of the strategy example and the transaction record of the day to obtain warehouse holding data and strategy example fund data; s4: updating the funds of the complex disc and the holding warehouse to a database, calculating each funds of the account and updating, and recording corresponding funds snapshot after all funds are calculated; s5: and finally, carrying out daily cleaning. And the handling fees, the filling and the depletion of the flat warehouse and the like are all cleared, so that the normal funds of the next trade day are ensured.

The muji system performs global market counter routing forwarding through a unified channel gateway. The passwords of the fund account are stored in the database in a symmetrical encryption mode. The online transaction process monitoring of the transaction platform comprises trade-server monitoring, trade-schedule monitoring, trade-desk-server monitoring, futures transaction container, futures online container, foreign exchange transaction container, foreign exchange online container and the like. The transaction platform adopts a log4j2 asynchronous log printing mode, and log monitoring alarm contents comprise ctp client monitoring (ctp interface overtime and system abnormality), order management alarm, algorithm transaction alarm, snapshot data alarm, transaction unit processing abnormality, database monitoring, user login failure, all forward normal message notification, number of times of self-success has been reached, number of times of frequent report and withdrawal has been reached, number of times of large report and withdrawal has been reached, and wind funding has been reached.

And II, a data platform is used for inputting global market quotation data in real time, synthesizing K lines and indexes through stream processing, pushing real-time quotation in real time, providing a data service interface and cleaning and processing historical quotation data as shown in figure 11.

As shown in fig. 10, the flow of implementing multipath CTP market access processing by the data platform is as follows: accessing basic quotation data from a data source, and using different network lines by a receiving client to ensure the stability of the data to the maximum extent; each client sends the received quotation to a quotation processing service through netty; monitoring whether each server has quotation, and if one or more clients do not have quotation, sending out alarm information; monitoring basic surface information sent by a server, judging whether the basic surface information is consistent or not, and sending alarm information if the basic surface information is inconsistent; and correspondingly processing the given alarm information, such as processing the application problem of the client, replacing the front of a future company and the like. And carrying out quotation optimization on quotation data received from different clients according to the time priority and the transaction priority, wherein the data platform uses the quotation data after the optimization to carry out subsequent quotation processing.

And (3) accessing IB quotations: starting two IB Gateway to acquire market data, respectively receiving the market data of the two Gateway by the two IB-clients, screening the market in the market processing service, selecting the market of the main market source, carrying out subsequent K line synthesis and other processing. IB multi-activity quotation monitoring flow: the timing task is 1 time per second, the time of avoiding the just-opened time and the time of starting the application are considered, when the main market or the standby market exceeds 5 seconds and no market exists, the warning is carried out, and the warning control frequency (only one warning is sent out in five minutes); if the main quotation source is abnormal and the standby quotation source is normal, the quotation source needs to be switched, the main quotation source in the cache is set as the code of the standby quotation source, and the code of the main quotation source is stored in the cache (redis) after the switching.

The real-time K line synthesis process comprises the following steps: and (3) receiving the selected market of the Tick, checking the transaction time, performing the following steps at the transaction time, sending the Tick data to mq according to whether the market data is Zheng Shang processed, acquiring the Tick of the last cache, acquiring a K line set in the cache, synthesizing a low-frequency K line according to a low-frequency K line period, synthesizing a high-frequency K line according to a high-frequency K line period, asynchronously warehousing the Tick data, and updating a redis cache K line set. Real-time K line segmentation process: executing processing of a timed task every minute, acquiring contract information, segmenting K lines in a cache according to whether K line segmentation time is reached, traversing subscribed contracts, executing a subsequent K line segmentation program by multithreading, acquiring a contract transaction time period, acquiring a K line set cache corresponding to the contract, calculating the accumulated transaction time number, traversing a K line period, judging whether the K lines should be segmented, backing up the K lines in the cache if the K lines need to be segmented, emptying the cache, asynchronously warehousing the K lines, and pushing the K lines to a policy container.

The quotation subscription process: storing the subscription push relation between the contract and the container by using a td_substriction_relation table; initializing the relation between a subscription contract loaded from the new establishment and a container ip and port; monitoring subscription information through mq after starting; maintaining the subscription relation (including subscription and unsubscribe) in the memory after receiving the subscription message; the subscription relationship is persisted to the database. And (3) a market push process: judging whether to push processInBar when synthesizing the K line; judging whether pushing the bar is completed or not by cutting the K line; and if bar needs pushing, pushing the bar to the target container through tcp connection according to the subscription relation.

The index synthesis process comprises the following steps: inquiring contract variety data; receiving a tick market from the mq, judging whether the first tick is the first tick, and setting the timing time t1 as a time stamp of the tick if the first tick is the first tick; comparing the first pen with the timing time t 1; generating an exponential tick according to the buffered tick data if the difference is greater than 500 ms; buffering the tick data, synthesizing a K line according to the index contract tick, warehousing the K line data, and pushing a strategy platform by the K line data.

The global market quotation processing process of the data platform comprises the following steps: accessing basic market data of all markets worldwide from different data suppliers; pushing the basic quotation data to the mq cluster, and landing the basic quotation data; each market processing application subscribes basic market data and processed market data from the mq cluster to perform market processing, such as variety index synthesis, cross-period, cross-variety arbitrage market processing, cross-market derived data processing and the like; and pushing the processed market data to the mq cluster, and landing the processed market data.

The derived data embody the change rule between the indexes or basic contract quotations of two or more varieties, and the purpose of processing the derived data is to provide market support for arbitrage trading, and mainly comprises three forms: cross-period arbitrage derivative data: reflecting the change of price gap between different delivery periods of the same commodity in the same market; cross-breed arbitrage derivative data: reflecting the variation of price gap between different commodity varieties in the same market (generally refers to varieties with correlation, such as corn and corn starch, soybean oil and soybean meal, etc.; data derived across market arbitrage: reflecting the change of price differences of the same commodity variety in different markets, wherein the data platform derived data processing flow is as follows: creating derivative data definitions, warehousing the derivative data definitions, calculating a public transaction time period, subscribing basic contracts contained in the derivative data definitions in the mq cluster, acquiring basic contract market conditions, calculating price differences by using a rule engine Aviator according to a calculation formula in the derivative data definitions, generating derivative data tics, combining K lines of the derivative data tics, cutting the K lines of the derivative data tics, and pushing the K lines of the derivative data to a policy container to drive policy operation.

History data cleaning process: loading a cleaning target after starting, and inquiring variety and contract data; cleaning the tick data according to transaction days; and cleaning the K line data according to the transaction date, and writing the abnormal quotation information into the log file. History data processing process: loading variety and contract information; loading basic contract tics to generate index tics; generating a high-frequency index K line and a low-frequency index K line by using an index tick; and (5) warehousing the K line.

The data platform provides other platform data query services through the dubbo interface. High concurrency, high efficiency and real-time processing, high availability, stable system, reasonable data storage structure and data interaction design, and data processing performance and accuracy.

III, strategy platform: as shown in fig. 12 and 13, muji-manager is a portal for the entire wood chicken system, and all front-end access requests are forwarded to the background service via muji-manager. The policy platform contains two main subsystems: 1. offline policy production management; 2. and (5) performing management on the online strategy. The offline policy production mainly comprises 3 sub-modules: 1. policy production module (the main functions are flow control, report presentation, list query, etc.); 2. the strategy loop-back optimization module (the main functions are the creation and operation of strategy loop-back tasks, the scheduling of parallel computing tasks and the injection of strategy into a loop-back container); 3. the factor calculation module establishes a network link with the return container during the online policy operation, receives a policy signal from the return container, calculates a policy evaluation factor and a score (single curve) according to the signal, and calculates the comprehensive score (comprehensive curve) of the policy on a plurality of varieties after all the policy instances are operated.

Setting a dimension rule in a strategy platform: a series of rules are defined in a certain dimension, and indexes such as MACD, MA, RSI and the like are a set of rules, and meta-rules in different dimensions are intersected with each other in a combination way, so that better descriptive capability is generated. Wiener: jian Shandian A dimension is an atomic logic (not subdivided) describing the market, belonging to a specific dimension rule, having multiple/empty directions, different elements under the same dimension rule being mutually exclusive, all elements adding up all market states over the time sequence. An image: and stacking the permutation and combination of different periods through a plurality of dimension elements. To describe the market more precisely from multiple dimensions, the formula: image = dimension (element) [ period ] ×dimension (element) [ period ] ×.

The policy platform includes a plurality of policy subunits: the system comprises an image subunit, a bin opening subunit, a bin leveling subunit, a fund management subunit, an air control subunit and the like, wherein each type of subunit can only form strategies with different types of subunits, a plurality of strategy subunits are arranged and combined to form a plurality of strategies, and the formula is as follows: the drawing subunit is divided into an open bin subunit, a flat bin subunit, a fund management subunit and an air control subunit.

The business process of the policy platform comprises the following steps: and D, dimension element management: uploading the uploaded element codes to a system, and supporting later maintenance; and (3) arranging dimension elements: selecting a plurality of dimension elements according to the dimension, and performing full permutation and combination to generate an image; image evaluation: a batch of images generated by one-time dimension arrangement are subjected to back testing, evaluation and screening, and images and parameters meeting the conditions are found; policy subunit management: uploading codes of the strategy subunits to a system and supporting later maintenance; strategy arrangement: the selected strategy subunits are fully arranged to generate strategies; strategy screening: according to a batch of strategies generated by primary strategy arrangement, carrying out one-by-one back measurement and screening according to a specified screening rule, and taking the strategies meeting the conditions as effective strategies to carry out warehousing operation; policy adaptation: aiming at the warehousing strategy passing the screening, carrying out parameter adaptation to find the best transaction variety corresponding to the effective parameter set; the method comprises the following steps of: setting a strategy instance (strategy + parameter + variety) meeting the conditions as to-be-online; policy online: and pushing the strategy to be online from the strategy library to online, and entering a real disc or simulated disc running state.

The strategy platform mainly comprises a product/optimize/factor/muji-strategy-api. In order to avoid circular dependence, each module interaction principle is that an upstream module calls a downstream module to adopt a dobbo service, and the downstream module informs the upstream module of adopting an MQ mode. The product module is responsible for image evaluation main flow control, service inquiry and metadata management, and when the image return time is involved in the image evaluation process, the image return is realized by invoking optimize through dubbo; the Optimize is responsible for the return measurement of images and strategies, optimization, task scheduling of return measurement, acquisition of containers by dubbo call infra structure during return measurement, strategy injection and strategy initial instruction sending; the Infrastructure is responsible for interacting with the policy container, including the management of the online policy container and the offline policy container; muji-strategy-api: taking charge of strategy online and starting and stopping of online strategy; factor: is responsible for accepting orders from the offline container corelocker at the time of the return test, calculating policy factors and scores, and notifying the optimize or product module of the completion of calculation by sending an MQ message.

And (5) returning the container cluster: the concurrent running of the back-testing task of the back-testing container is large in resource consumption, a cluster mode is adopted, the cluster comprises 6 machines, and each node deploys a container process to support horizontal expansion.

Factor calculation clusters: the resources are occupied, the single machine is deployed in multiple instances, 4 services are deployed in one of the single machine, 2 services are deployed in the other one of the single machine, and the number of the services and the number of the return containers are 1:1.

On-line policy container cluster: and an online strategy is operated, single-node multi-container is supported, and horizontal expansion is supported.

Image evaluation process:

the Manager receives an image evaluation task starting request and transmits the request to the product module through a dubbo call;

the product starts image evaluation and optimizing step by step, and each image of each layer needs to call optimize for back measurement;

the Optimize receives the image return request, creates an image return task, and starts the image return task by parameter frying (permutation and combination);

starting the loop-back task comprises the following steps: acquiring available policy container resources, injecting policies into a container, and sending a starting instruction to the container; when the strategy in the container back measurement enters the operation, an order signal is sent to a Factor, and the Factor receives the order to start calculating the strategy performance; after the Factor calculates the policy performance, a report of completion of the test is sent back to the Optimize, and the Optimize performs task ending processing according to the report, including modifying task state bodies, and releasing container resources.

As shown in fig. 15, policy container-asynchronous event driven process: different market signal driving modes are adopted aiming at different environments: if the environment is a return environment, directly reading a local Sqlite file through a monitor in the container to send quotation, and if the environment is an online container, subscribing quotation from the Data-Source; asynchronous event delivery: quotation/signals are circulated in the policy container in an asynchronous event mode; the online strategy supports the market subscription and driving of the complete Bar, currentBar and Tick full period/full variety; the internal reading history market data of the strategy adopts a local memory cache, so that a large number of IO operations are reduced, and the operation time of the strategy is compressed to microsecond level; after the strategy in the container generates the signal, whether the signal is back-measured or on-line is judged according to the environment variable, if the signal is a back-measured task, the signal is sent to the Factor module for performance calculation, and if the signal is an on-line container, the signal is directly sent to the transaction module and then forwarded to the transaction market.

The parallel computing task scheduling process of the strategy module comprises the following steps: a request submitted by a user corresponds to a father task, the father task does not directly enter a task waiting queue, but is split into atomic tasks with relatively uniform granularity, and enters the waiting queue to wait for being scheduled; the task waiting queues are ordered according to task priorities set by users, the tasks with high priorities are close to the Head and are operated first, and the tasks with the same priorities are ordered according to time sequence; the scheduling module circularly acquires tasks corresponding to the Head of the waiting queue, acquires available computing resources, and if available idle resources exist, the Head tasks are moved to a task list in operation and enter an operation state; the running task is removed from the running task list.

In particular, a task scheduling module of the optimal subsystem manages the operation and resource allocation of atomic tasks, a tree structure is constructed in the task creation process, a root node represents one task seen by a user, and leaf nodes are atomic tasks; the atomic tasks are ordered according to the priority, enter a task waiting queue, a task scheduling module polls the head of the waiting queue at regular time, if the available computing resources are found, the resources are allocated to the tasks, the head tasks are moved to an operation list, and the task back-testing operation is started; the optimization receives the notice from the Factor subsystem, starts the processing flow of task completion, if the task is successfully ended, releases resources, notifies the task to schedule and starts the next task, and other business processing flows exist; if the task fails, the resources are released, the task is retried for 3 times, if the retry fails for 3 times, the task is considered to truly fail in the service, then a failure processing flow is entered, and the failure of the atomic task affects the state of the father task, including the state of the root node of the task tree and the state of the brother task.

The strategy module forms a tree structure of the portrait through dimension arrangement, and each node on the tree structure carries out the return evaluation of the whole market in a layer-by-layer optimizing mode. And (3) a step-by-step optimization process of image evaluation: the image evaluation is performed for each node of the multi-layer tree structure of the image generated by one-time dimension element arrangement; each node represents an image, each image has a score and an optimal parameter corresponding to the score, and only images with scores greater than 0.6 are considered to be valid images in service, and the parameter with the highest score is taken as the optimal parameter of the image. In order to reduce the calculation amount, the optimal parameters corresponding to the upper node (image) are inherited to the next image, and the lower image is subjected to the explosion (parameter arrangement-Cartesian product) back evaluation on the dimension elements of the layer on the basis of the optimal parameters of the upper image. The evaluation criteria for the images of different levels are different, and for one image, the second image is the loss per unit time length in terms of net profit. In order to prevent the problem of over-fitting of the generated effective images, a comprehensive curve of all varieties is adopted for carrying out a platform in the evaluation process, and meanwhile, a parameter plains algorithm is also applied to parameter optimization, so that the evaluation of the effective images can participate in strategy arrangement.

The cluster performance optimization scheme has the technical effects that:

and through a task scheduling engine and a resource pool management module, dynamically managing and distributing resources, splitting one large task into small callback tasks, and simultaneously running a plurality of small callback tasks on different machines. And 6 servers are adopted to form a cluster for parallel back measurement, and the throughput of the cluster is 185.53 kilobars/second. Six years of data are used, a period is 15 minutes, only one image layer is provided, each image is exploded into 492 groups of parameters, full variety image evaluation is carried out on 6 servers, 28536 examples take 7.2 minutes on average, production is carried out according to the result evaluation by using 6 years of data, and 570.72 ten thousand strategies can be tested every day.

The technical effect of the step-by-step optimizing algorithm scheme is as follows:

the method comprises the steps of arranging a plurality of dimension elements into a multi-level tree structure in a permutation and combination mode, evaluating step by step from the root of the tree, searching for optimal parameters, and transmitting the optimal parameters to a next-level image. This has the advantage of reducing a number of invalidation operations, saving computational resources, taking a 3-cycle double-average line as an example, each node being 100 sets of parameters, and after permutation, the first layer is 2 nodes, the second layer is 4 nodes, and the third layer is 8 nodes.

The full-arrangement mixing-level optimizing operation amount is adopted: 100 x 2+ (100 x 4) +. 100 x 8 () = 8040200;

the step-by-step optimizing operand is adopted: 100×2+100×4+100×8=1400;

in comparison, the total-arrangement mixed-level optimizing operand is 5700 times of that of the step-by-step optimizing, and the system operand is only 1/5700 of that of the total arrangement after the step-by-step optimizing, so that the improvement effect is obvious.

Under-fitting solution: under fitting means that the general feedback sample data volume is too small, so that the fitting strategy cannot meet the requirement, the error is larger, the solution is a large time range, the large sample feedback is realized, the current time range for domestic futures market feedback is 2012-01-01 to 2019-3-31 years, and the data of nearly 7 years is obtained; the other method is to carry out the back test aiming at the whole market, such as the image evaluation and the strategy screening of domestic futures, and 29 varieties with better liquidity are selected for the back test, so that the problem of under fitting can be effectively avoided.

Solution of overfitting: overfitting means that the strategy overfits historical data without generalization capability, and the solution is to limit the maximum layer number of the strategy image to only 3 stages in the image evaluation process, limit parameter islands by adopting plain parameters, and separately test the strategy in the sample and the strategy outside the sample, wherein the training strategy in the sample and the return test outside the sample are used for verifying the validity of the strategy generated in the sample.

Parameter plains algorithm: determining a target factor: from the currently available evaluation factors, determining a factor as a basic evaluation factor of a plains algorithm (such as net profit, annual income ratio and the like); determining neighbor length: the neighbor length is a positive integer. The default value is 2, and the business personnel can modify the range of the parameter set used for determining the constructed parameter plain. Let a certain parameter set have M parameters, find the neighbor parameter set (+n-N step size of every 1 parameter, when taking the neighbor parameter of this parameter, other parameters in the parameter set are fixed first, every 1 parameter has 2N neighbor parameter sets and target factor value (excluding this parameter itself), M parameters have m×2n neighbor parameter sets, finally add parameter itself, then there are m×2n+1 parameter sets and corresponding target factor value altogether. For example: in the double-average line strategy, the Fast parameter group (1-50) corresponds to 50 parameters, and the step length is 1; the low parameter group (0-200) corresponds to 21 parameters, and the step length is 10; when the neighbor length n=2 is set and the parameter set fast=6 and slow=40 is selected, the parameter set ranges are (6, 20), (6, 30), (6,40), (6,50), (6, 60), (4, 40), (5, 40), (7, 40) and (8, 40).

The strategy production process adopts a multi-level report form: because a large amount of intermediate temporary data is generated in the process of policy production, a large amount of storage space is occupied, in order to save consumption of storage resources, reports of different levels are adopted according to different business requirements, and a background program stores data according to the levels of the reports. Storage and calculation order: simple report > summary report > score report; for low-level reports, if the user needs to see more data, the report data of high level can be seen by using the report expanding function.

Strategy comprehensive scoring formula: f=n+m/100;

where n=annual non-lever gain/maximum withdrawal (representing profitability, the higher the gain the stronger the profitability at the same risk level) m= (PR-Q)/R (representing effectiveness, P-win, R-loss, q=1-P), and the composite score f >0.6 score is considered as an effective strategy.

Stock multi-factor calculation:

net value performance:

annual yield: (netValue (end) -1)/(250/length (netValue)), netValue (end) being the net value sequence end value, length (netValue) being the net value sequence length;

the kama ratio: annual rate of return/historical maximum withdrawal;

ratio of summer: (annual rate of return-0.03)/annual rate of fluctuation; annual fluctuation rate=std (r_i) ×v 250.

IC (information coefficient) =corr (F (t), R (t+1)), and the correlation coefficient (reflecting the prediction ability of the factor for future gain) of the current-period factor value sequence F (t) and the next-period gain-rate sequence R (t+1) is a sequence value.

IC mean, the average value of IC sequences;

IC std. standard deviation of IC sequences;

Ann.ICIR:

Ann.ICIR＝(mean(IC))/(std(IC))√C

wherein C is the number of times of bin adjustment in one year.

Hand change rate turn over, i.e. the number of tickets is called out by the ith group in the next period/the total number of the ith group in the current period;

annual change rate Ann. Turnover: change rate Turnover annual number of bin adjustments C.

As shown in fig. 18, automated production of policies: creating an overall scheduling task in the product, receiving page data, storing mongo, then starting dimension element arrangement, after the task is successful, sending an MQ notification to the product module, and if the execution of a certain step fails, automatically constructing the overall task to be stopped. As a plurality of steps are arranged in the middle of manual operation from uploading the dimension element to finally generating the effective strategy, the operation is complicated, a line of functions can be added in the automatic production process of the strategy, and the dimension element arrangement, image evaluation, strategy arrangement and strategy screening processes are spliced into one process, so that the operation steps are simplified.

The policy platform employs a machine learning algorithm to calculate the weight of the funding proportion for each policy in the portfolio. The purpose of the online portfolio is to select portfolio weights at each time period to maximize its final financial value.

As shown in fig. 19, in order to prevent excessive fitting during large-scale policy production (parallel computing) -policy loop test, long-period and full-market loop test is adopted when the policy is loop tested, so that the operation amount is huge, and a single machine cannot meet the requirement of loop test performance, so that a multi-machine parallel/single machine multi-thread loop test architecture is designed to support loop test of massive policy examples. And (3) a strategy loop-back task, wherein the generated strategy examples are too many, and the task is automatically split into sub-tasks if the capability upper limit of one machine is exceeded, and the splitting granularity of the sub-tasks is based on the upper limit of the strategy examples which are operated by a single machine at the same time.

Technical Effect control example:

the scheme is as follows: by adopting single machine multithreading, each machine runs 200 threads simultaneously, shared data and concurrent locks are not arranged among each thread, and by adopting a local memory to cache common data and a K-line quotation optimization algorithm, a large number of IO operations are reduced, the CPU occupies nearly 100% after a task is started, time slices are concentrated in user time, and the performance of the multi-core CPU is utilized to the greatest extent.

Results: the method is characterized in that six-year data and one-minute data of the domestic futures market are used before, parallel return testing is carried out on 6 servers, the average time of a single instance is 0.5 seconds, the average throughput of a single machine is 47.83 kilobars/second, and compared with a return platform Xquant of the main flow in the industry developed by a large merchant, the single machine performance is 12.5 times of that of the Xquant.

As shown in fig. 20, the large-scale policy operates (linear expansion): policies that operate simultaneously with a single policy container are limited, and generally within 1000 policies can guarantee flow operation, assuming that an average of 10 tens of thousands of funds can be allocated to a policy, this limits the platform to only 1 million scale funds (1000 x 10 tens of thousands=1 million), but if 100 tens of thousands of policies are to be operated, managing 1000 million funds would necessitate that the policy container support clusters and have the ability to be linearly extended horizontally.

While the present invention has been described in detail with reference to the embodiments, the present invention is not limited to the above-described embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art, and the present invention shall also be considered as the scope of the present invention.

Claims (6)

1. A market-wide multi-variety financial resource management system based on an intelligent policy platform, comprising:

the transaction platform is responsible for receiving the newspaper command, executing the algorithm transaction and interfacing the market newspaper function, and the transaction newspaper flow of the transaction container in the transaction platform is as follows: s1: after the transaction signal reaches the transaction container, the signal initialization and the strategy bin association are realized through logic processing of the signal management module; s2: after the signal is dequeued from the signal management module, the signal enters the order management module for filtering processing to generate a mother bill; s3: the master list entering algorithm module processes the parallel and parallel instruction through the algorithm route, and then routes the parallel and parallel instruction to the corresponding list splitting algorithm to split the list, and after all the hand numbers split the list, a signal for finishing the list splitting is generated for the master list to the algorithm post module; s4: the sub-bill generated in the bill disassembling process is directly sent to a fund checking module to carry out wind funding and fund checking logic, and the sub-bill is sent to a market route after checking, otherwise, the fund checking refuses the bill; 5: the transaction routing module defaults to send a request to the transaction channel gateway in a java plug-in mode, and sends the request to a corresponding transaction market through routing; s6: the return sent back from the market is received by the client side built in the transaction channel, then the basic service of container management is called to send the return back to the transaction container, the return entering the transaction container is received by the sub-bill state monitor and triggers corresponding state circulation, the state processing of returning the deposit, removing the bill or refusing the bill is carried out, and finally the state promotion of fund calculation and mother bill is triggered; in the transaction report flow S2, the order management module performs three-layer filtering: processing signals of the type under the main force mapping to obtain corresponding main force contracts, aligning bin positions, processing logic of automatic alignment or non-automatic alignment of bin positions to correct the number of single hand of the newspaper, carrying out secondary confirmation of the strong flat type, and if the number of single hand of the newspaper is not consistent with the bin position of a real disk, re-initializing the number of single hand of the newspaper to finally generate a mother bill; after the transaction report flow S6, the transaction report triggers to perform fund calculation: calculating layer by layer in the form of java plugins, namely firstly holding the warehouse in the dimension of the policy instance and fund in the dimension of the warehouse, then carrying out fund in the dimension of the policy instance, and finally carrying out fund in the dimension of the account; after the transaction report flow S6, triggering the corresponding action of the master bill, pushing the state of the master bill along with the corresponding action, and if the master bill is ended, feeding back the result to the signal management module to complete the pushing of the signal state;

The data platform is responsible for service quotation access and production of related derived quotations;

the strategy platform is responsible for the online operation of strategies for strategy production and high performance, comprises an offline strategy production management subsystem and an online strategy operation management subsystem, and adopts a machine learning algorithm to calculate the weight of the fund proportion of each strategy in the combination, and selects investment combination weights in each period so as to maximize the final benefit; the policy platform comprises an online policy container cluster, a policy platform and a policy platform, wherein the online policy container cluster supports single-node multi-container and horizontal expansion; the online operation link adopts a CEP event processing engine, and is constructed on an Apama container to circulate core business;

the offline policy production management subsystem includes 3 sub-modules: the strategy production module is used for realizing the functions of flow control, report display and list query; the strategy return testing optimizing module is used for realizing the establishment and operation of strategy return testing tasks, the scheduling of parallel computing tasks and the injection of strategies into a return testing container; the return container module is used for running the strategy during return, sending quotation to the strategy, calculating funds and holding a warehouse during running the strategy and providing strategy API common strategy calling; during the operation of the on-line strategy operation management subsystem, the factor calculation module establishes a network link with the return measurement container, receives strategy signals from the return measurement container, calculates strategy evaluation factors and a grading single curve according to the signals, and calculates comprehensive grading comprehensive curves of strategies on a plurality of varieties after all strategy examples are operated;

The strategy platform comprises produce, optimize, factor and strategy-api modules, and the interaction principle of each module is that an upstream module calls a downstream module to adopt dobbo service, and the downstream module informs the upstream module of adopting an MQ mode; the product module is responsible for image evaluation main flow control, service inquiry and metadata management, and the image return time is related in the image evaluation process, and the image return is realized by dubbo calling optimize; the optimize is responsible for the return measurement of images and strategies, optimization, task scheduling of return measurement, acquisition of containers by dubbo call infracuction, strategy injection and strategy initial instruction sending; the Infrastructure is responsible for interacting with the policy container, including the management of the online policy container and the offline policy container; the strategy-api is responsible for online policy, and starting and stopping of online policy; the factor is responsible for receiving orders from an offline container corelear during back measurement, calculating strategy factors and scores, and notifying an optimize or a product module by sending an MQ message after calculation;

the policy platform includes a cluster of retested containers: the cluster comprises 6 machines, each node deploys a container process to support horizontal expansion; the policy platform includes a factor computation cluster: the single machine multi-instance deployment, wherein one deploys 4 services, the other deploys 2 services, and the number of services and the number of the callback containers are 1:1.

2. The intelligent policy platform-based market-wide multi-variety financial resource management system of claim 1, wherein the policy platform image evaluation process: the Manager receives an image evaluation task starting request and transmits the request to the product module through a dubbo call; the product starts image evaluation and optimizing step by step, and each image of each layer needs to call optimize for back measurement; the Optimize receives the image return request, creates an image return task, realizes parameter frying through permutation and combination, and starts the return task.

3. The intelligent policy platform-based market-wide multi-variety financial resource management system of claim 2, wherein the policy platform initiation loop tasks comprise: acquiring available policy container resources, injecting policies into a container, and sending a starting instruction to the container; when the strategy in the container back measurement enters the operation, an order signal is sent to a Factor, and the Factor receives the order to start calculating the strategy performance; after the Factor calculates the policy performance, a report of completion of the test is sent back to the Optimize, and the Optimize performs task ending processing according to the report, including modifying task state bodies, and releasing container resources.

4. The intelligent policy platform based all-market multi-variety financial resource management system of claim 3, wherein the policy container of the policy platform is asynchronous event driven process: if the environment is the return environment, directly reading a local Sqlite file through a monitor in the container to send quotation; if the online container is an online container, subscribing to quotations from the Data-Source; asynchronous event delivery: quotation/signals are circulated in the policy container in an asynchronous event mode; the online strategy supports the market subscription and driving of the complete Bar, currentBar, tick full period and full variety; the strategy internal reading history market data adopts a local memory cache to reduce a large number of IO operations, and the strategy running time is compressed to microsecond level; after the strategy in the container generates the signal, whether the signal is back-measured or on-line is judged according to the environment variable, if the signal is a back-measured task, the signal is sent to the Factor module for performance calculation, and if the signal is an on-line container, the signal is directly sent to the transaction module and then forwarded to the transaction market.

5. The intelligent policy platform based all-market multi-variety financial resource management system of claim 4, wherein the under-fitting solves: the method comprises the steps of carrying out back testing aiming at the whole market, selecting a variety with better fluidity for carrying out back testing, and avoiding the problem of under fitting through image evaluation and strategy screening; solution of overfitting: in the process of image evaluation, the maximum layer number of the strategy image is limited to 3, plain parameters are adopted, parameter islands are limited, and in addition, a separate test between the inside and the outside of the sample is adopted, the strategy is trained in the sample, and the outside of the sample is tested in a return mode to verify the validity of the strategy generated in the sample.

6. The intelligent policy platform based multi-breed financial resource management system of the world wide market of claim 5, wherein the automated production process of policies: creating an overall scheduling task in the product, receiving page data, storing mongo, then starting dimension element arrangement, after the task is successful, sending an MQ notification to the product module, and if the execution of a certain step fails, automatically constructing the overall task to be stopped.